o &j}@sdZddlmZmZmZddlmZmZddlZ ddl Z ddl Z ddl Z ddlmZddlmZdd lmZdd lmZdd lmZdd lmZd dZd[ddZddZddZGdddejZ GdddejZ!Gddde!Z"Gddde"Z#Gddde"Z$Gd d!d!e!Z%Gd"d#d#e!Z&Gd$d%d%e!Z'Gd&d'd'e!Z(Gd(d)d)e!Z)Gd*d+d+e!Z*Gd,d-d-e!Z+e ,eGd.d/d/e-Z.Gd0d1d1e.Z/Gd2d3d3e.Z0Gd4d5d5e.Z1Gd6d7d7e1Z2Gd8d9d9e1Z3Gd:d;d;e.Z4Gdd?d?e.Z6Gd@dAdAe.Z7GdBdCdCe.Z8e j9dDdEdF G  Hd\dIdDZ:e j9dJdKdF G  Hd\dLdJZ;e j9ddMdF G  P   Hd]dRdSZ deterministic random_stateZ derive_rng_)r?augrrr_to_deterministics     rEcsReZdZdZ    dfdd Zdd Zd d Zd d ZddZddZ Z S) BlendAlphaa+ Alpha-blend two image sources using an alpha/opacity value. The two image sources can be imagined as branches. If a source is not given, it is automatically the same as the input. Let ``FG`` be the foreground branch and ``BG`` be the background branch. Then the result images are defined as ``factor * FG + (1-factor) * BG``, where ``factor`` is an overlay factor. .. note:: It is not recommended to use ``BlendAlpha`` with augmenters that change the geometry of images (e.g. horizontal flips, affine transformations) if you *also* want to augment coordinates (e.g. keypoints, polygons, ...), as it is unclear which of the two coordinate results (foreground or background branch) should be used as the coordinates after augmentation. Currently, if ``factor >= 0.5`` (per image), the results of the foreground branch are used as the new coordinates, otherwise the results of the background branch. Added in 0.4.0. (Before that named `Alpha`.) **Supported dtypes**: See :func:`~imgaug.augmenters.blend.blend_alpha`. Parameters ---------- factor : number or tuple of number or list of number or imgaug.parameters.StochasticParameter, optional Opacity of the results of the foreground branch. Values close to ``0.0`` mean that the results from the background branch (see parameter `background`) make up most of the final image. * If float, then that value will be used for all images. * If tuple ``(a, b)``, then a random value from the interval ``[a, b]`` will be sampled per image. * If a list, then a random value will be picked from that list per image. * If ``StochasticParameter``, then that parameter will be used to sample a value per image. foreground : None or imgaug.augmenters.meta.Augmenter or iterable of imgaug.augmenters.meta.Augmenter, optional Augmenter(s) that make up the foreground branch. High alpha values will show this branch's results. * If ``None``, then the input images will be reused as the output of the foreground branch. * If ``Augmenter``, then that augmenter will be used as the branch. * If iterable of ``Augmenter``, then that iterable will be converted into a ``Sequential`` and used as the augmenter. background : None or imgaug.augmenters.meta.Augmenter or iterable of imgaug.augmenters.meta.Augmenter, optional Augmenter(s) that make up the background branch. Low alpha values will show this branch's results. * If ``None``, then the input images will be reused as the output of the background branch. * If ``Augmenter``, then that augmenter will be used as the branch. * If iterable of ``Augmenter``, then that iterable will be converted into a ``Sequential`` and used as the augmenter. per_channel : bool or float or imgaug.parameters.StochasticParameter, optional Whether to use the same factor for all channels (``False``) or to sample a new value for each channel (``True``). If this value is a float ``p``, then for ``p`` percent of all images `per_channel` will be treated as True, otherwise as False. seed : None or int or imgaug.random.RNG or numpy.random.Generator or numpy.random.BitGenerator or numpy.random.SeedSequence or numpy.random.RandomState, optional See :func:`~imgaug.augmenters.meta.Augmenter.__init__`. name : None or str, optional See :func:`~imgaug.augmenters.meta.Augmenter.__init__`. random_state : None or int or imgaug.random.RNG or numpy.random.Generator or numpy.random.BitGenerator or numpy.random.SeedSequence or numpy.random.RandomState, optional Old name for parameter `seed`. Its usage will not yet cause a deprecation warning, but it is still recommended to use `seed` now. Outdated since 0.4.0. deterministic : bool, optional Deprecated since 0.4.0. See method ``to_deterministic()`` for an alternative and for details about what the "deterministic mode" actually does. Examples -------- >>> import imgaug.augmenters as iaa >>> aug = iaa.BlendAlpha(0.5, iaa.Grayscale(1.0)) Convert each image to pure grayscale and alpha-blend the result with the original image using an alpha of ``50%``, thereby removing about ``50%`` of all color. This is equivalent to ``iaa.Grayscale(0.5)``. >>> aug = iaa.BlendAlpha((0.0, 1.0), iaa.Grayscale(1.0)) Same as in the previous example, but the alpha factor is sampled uniformly from the interval ``[0.0, 1.0]`` once per image, thereby removing a random fraction of all colors. This is equivalent to ``iaa.Grayscale((0.0, 1.0))``. >>> aug = iaa.BlendAlpha( >>> (0.0, 1.0), >>> iaa.Affine(rotate=(-20, 20)), >>> per_channel=0.5) First, rotate each image by a random degree sampled uniformly from the interval ``[-20, 20]``. Then, alpha-blend that new image with the original one using a random factor sampled uniformly from the interval ``[0.0, 1.0]``. For ``50%`` of all images, the blending happens channel-wise and the factor is sampled independently per channel (``per_channel=0.5``). As a result, e.g. the red channel may look visibly rotated (factor near ``1.0``), while the green and blue channels may not look rotated (factors near ``0.0``). >>> aug = iaa.BlendAlpha( >>> (0.0, 1.0), >>> foreground=iaa.Add(100), >>> background=iaa.Multiply(0.2)) Apply two branches of augmenters -- ``A`` and ``B`` -- *independently* to input images and alpha-blend the results of these branches using a factor ``f``. Branch ``A`` increases image pixel intensities by ``100`` and ``B`` multiplies the pixel intensities by ``0.2``. ``f`` is sampled uniformly from the interval ``[0.0, 1.0]`` per image. The resulting images contain a bit of ``A`` and a bit of ``B``. >>> aug = iaa.BlendAlpha([0.25, 0.75], iaa.MedianBlur(13)) Apply median blur to each image and alpha-blend the result with the original image using an alpha factor of either exactly ``0.25`` or exactly ``0.75`` (sampled once per image). rNF deprecatedc stt|j||||dtj|ddddd|_|dus#|dus#Jdtj||jddd|_ tj||jd dd|_ t |d |_ d |_ dS) Nseedr&rCrBfactorrrT value_rangetuple_to_uniformlist_to_choicepExpected 'foreground' and/or 'background' to not be None (i.e. at least one Augmenter), but got two None values.r<defaultr> per_channelr)superrF__init__iaphandle_continuous_paramrLr handle_children_listr&r<r>handle_probability_paramrUepsilon) selfrLr<r>rUrKr&rCrB __class__rrrWs*    zBlendAlpha.__init__cCsPt||||\}}|j}|}t|} tdd|D} |d} |jj| | dd} |jj| | f| dd} t |D]i\}}| |dkr[t|dkrP|dnd}| |d|f}n| |j dkrh| |dfnd}t |}|dk}|j durt|j ||j |||jd|j |<|D]}|jd kr|r|n|}t||j||j|<qq<|S) NcSs$g|]}t|dkr|dndqS)r r)len).0r$rrr sz.BlendAlpha._augment_batch_..r rrCrr!r7images)rAcolumnsget_rowwise_shapesr`r3Z duplicaterU draw_samplesrL enumeraterr(averagerer9r\r&getattr attr_namevalue)r]r@rCr:r;batch_fgbatch_bgrfshapes nb_imagesZnb_channels_maxZrngsrUalphasrr$Z nb_channelsalphas_iZ alphas_i_avgZ use_fg_branchcolumnZ batch_userrr_augment_batch_sJ        zBlendAlpha._augment_batch_cCt|SNrEr]rrrrEzBlendAlpha._to_deterministiccCs |j|jgSz=See :func:`~imgaug.augmenters.meta.Augmenter.get_parameters`.)rLrUryrrrget_parameterss zBlendAlpha.get_parameterscCdd|j|jfDS)ASee :func:`~imgaug.augmenters.meta.Augmenter.get_children_lists`.cSg|]}|dur|qSrwrralstrrrrbz1BlendAlpha.get_children_lists..r<r>ryrrrget_children_listszBlendAlpha.get_children_listscCs*d}||jj|j|j|j|j|j|jfS)NzV%s(factor=%s, per_channel=%s, name=%s, foreground=%s, background=%s, deterministic=%s))r___name__rLrUr&r<r>rBr]patternrrr__str__s  zBlendAlpha.__str__rGNNFNNrIrI) r __module__ __qualname____doc__rWrurEr|rr __classcell__rrr^rrFs 2rFcsxeZdZdZdZdZeegZ   dfdd Zdd Ze d d Z e d d Z ddZ ddZ ddZddZZS)BlendAlphaMaska Alpha-blend two image sources using non-binary masks generated per image. This augmenter queries for each image a mask generator to generate a ``(H,W)`` or ``(H,W,C)`` channelwise mask ``[0.0, 1.0]``, where ``H`` is the image height and ``W`` the width. The mask will then be used to alpha-blend pixel- and possibly channel-wise between a foreground branch of augmenters and a background branch. (Both branches default to the identity operation if not provided.) See also :class:`~imgaug.augmenters.blend.BlendAlpha`. .. note:: It is not recommended to use ``BlendAlphaMask`` with augmenters that change the geometry of images (e.g. horizontal flips, affine transformations) if you *also* want to augment coordinates (e.g. keypoints, polygons, ...), as it is unclear which of the two coordinate results (foreground or background branch) should be used as the final output coordinates after augmentation. Currently, for keypoints the results of the foreground and background branch will be mixed. That means that for each coordinate the augmented result will be picked from the foreground or background branch based on the average alpha mask value at the corresponding spatial location. For bounding boxes, line strings and polygons, either all objects (on an image) of the foreground or all of the background branch will be used, based on the average over the whole alpha mask. Added in 0.4.0. **Supported dtypes**: See :func:`~imgaug.augmenters.blend.blend_alpha`. Parameters ---------- mask_generator : IBatchwiseMaskGenerator A generator that will be queried per image to generate a mask. foreground : None or imgaug.augmenters.meta.Augmenter or iterable of imgaug.augmenters.meta.Augmenter, optional Augmenter(s) that make up the foreground branch. High alpha values will show this branch's results. * If ``None``, then the input images will be reused as the output of the foreground branch (i.e. identity function). * If ``Augmenter``, then that augmenter will be used as the branch. * If iterable of ``Augmenter``, then that iterable will be converted into a ``Sequential`` and used as the augmenter. background : None or imgaug.augmenters.meta.Augmenter or iterable of imgaug.augmenters.meta.Augmenter, optional Augmenter(s) that make up the background branch. Low alpha values will show this branch's results. * If ``None``, then the input images will be reused as the output of the background branch (i.e. identity function). * If ``Augmenter``, then that augmenter will be used as the branch. * If iterable of ``Augmenter``, then that iterable will be converted into a ``Sequential`` and used as the augmenter. seed : None or int or imgaug.random.RNG or numpy.random.Generator or numpy.random.BitGenerator or numpy.random.SeedSequence or numpy.random.RandomState, optional See :func:`~imgaug.augmenters.meta.Augmenter.__init__`. name : None or str, optional See :func:`~imgaug.augmenters.meta.Augmenter.__init__`. random_state : None or int or imgaug.random.RNG or numpy.random.Generator or numpy.random.BitGenerator or numpy.random.SeedSequence or numpy.random.RandomState, optional Old name for parameter `seed`. Its usage will not yet cause a deprecation warning, but it is still recommended to use `seed` now. Outdated since 0.4.0. deterministic : bool, optional Deprecated since 0.4.0. See method ``to_deterministic()`` for an alternative and for details about what the "deterministic mode" actually does. Examples -------- >>> import imgaug.augmenters as iaa >>> aug = iaa.BlendAlphaMask( >>> iaa.InvertMaskGen(0.5, iaa.VerticalLinearGradientMaskGen()), >>> iaa.Sequential([ >>> iaa.Clouds(), >>> iaa.WithChannels([1, 2], iaa.Multiply(0.5)) >>> ]) >>> ) Create an augmenter that sometimes adds clouds at the bottom and sometimes at the top of the image. z either-orZ pointwiseNrIcstt|j||||d||_|dus|dusJdtj||jddd|_tj||jddd|_|j |j |j |j d|_ d|_ dS)NrJrRr<rSr>) keypointspolygons line_stringsbounding_boxesr) rVrrWmask_generatorr rZr&r<r>_MODE_POINTWISE_MODE_EITHER_OR _coord_modesr\)r]rr<r>rKr&rCrBr^rrrWds(    zBlendAlphaMask.__init__c Cstt||||\}}|j||}t|D]\}} |jdur/t|j||j|| |jd|j|<|jdur_|j|j} | j dd\} } | | | | } t|j|j|j|j| |jd|j|_|j dur|j |j } | j dd\} } | | | | } t|j |j |j |j | |jd|j |_ dD]%}t ||}|dur|||t |||t |||| |j|||<qq|S)Nrdrr )rrrr)rAr draw_masksrirer9r\ZheatmapsZarr_0to1r$_binarize_masksegmentation_mapsrrk_blend_coordinatesr)r]r@rCr:r;rnromasksrmaskr arr_height arr_widthZmask_binarizedZaugm_attr_nameZ augm_valuerrrrusZ              zBlendAlphaMask._augment_batch_c Cstt|}|jdkrtj||ftjd}n|jddkr#tj|ddnd}t|||f}t |dd}|dk}|S)Nrrr axisrr!) r(Z atleast_3drzerosr+r$rjiaimresize_single_imager4Zclip_) clsrrrZmask_3dZmask_rsZmask_avgZmask_arrZmask_arr_binarizedrrrrs   zBlendAlphaMask._binarize_maskcCst|}t|}t|}|}|}|}|j|jkr(|jks5nJd|j|j|jf|jdd\} } ||jkrt|t|ksZJdt|t|t|fg} t|||} | D]Y\} }}tt | d}tt | d}d|kr| krnn/d|kr| krnn#|||df}|j dkrt |nd}|dkr| |qd| |qd| ||fqdn|j dkrt |nd}|dkr|} n|} t jj| |jd }t||S) Na Expected number of coordinates to not be changed by foreground or background branch in BlendAlphaMask. But input coordinates of shape %s were changed to %s (foreground) and %s (background). Make sure to not use any augmenters that affect the existence of coordinates.rr zGot different numbers of coordinates before/after augmentation in BlendAlphaMask. The number of coordinates is currently not allowed to change for this augmenter. Input contained %d coordinates, foreground branch %d, backround branch %d.r.rr!)r$) augm_utilsZconvert_cbaois_to_kpsoisZ to_xy_arrayr$rr`zipintr(r#rrjrrZKeypointsOnImageZ from_xy_arrayZ invert_convert_cbaois_to_kpsois_)rZcbaoiZcbaoi_fgZcbaoi_bgZ mask_imagemodeZcoordsZ coords_fgZ coords_bgZh_imgZw_imgZ coords_augZsubgenZcoordZcoord_fgZcoord_bgZx_intZy_intrsr6Zmask_image_avgZ kpsoi_augrrrrsV      0   z!BlendAlphaMask._blend_coordinatescCrvrwrxryrrrrErzz BlendAlphaMask._to_deterministiccCs|jgSr{)rryrrrr|zBlendAlphaMask.get_parameterscCr})r~cSrrwrrrrrrbrz5BlendAlphaMask.get_children_lists..rryrrrrrz!BlendAlphaMask.get_children_listscCs&d}||jj|j|j|j|j|jfS)NzN%s(mask_generator=%s, name=%s, foreground=%s, background=%s, deterministic=%s))r_rrr&r<r>rBrrrrr#s  zBlendAlphaMask.__str__NNNNrIrI)rrrrrrZ_MODESrWru classmethodrrrEr|rrrrrr^rrs$f'/  Brcs6eZdZdZ    d fdd Zedd ZZS) BlendAlphaElementwisea Alpha-blend two image sources using alpha/opacity values sampled per pixel. This is the same as :class:`BlendAlpha`, except that the opacity factor is sampled once per *pixel* instead of once per *image* (or a few times per image, if ``BlendAlpha.per_channel`` is set to ``True``). See :class:`BlendAlpha` for more details. This class is a wrapper around :class:`~imgaug.augmenters.blend.BlendAlphaMask`. .. note:: Avoid using augmenters as children that affect pixel locations (e.g. horizontal flips). See :class:`~imgaug.augmenters.blend.BlendAlphaMask` for details. Added in 0.4.0. (Before that named `AlphaElementwise`.) **Supported dtypes**: See :class:`~imgaug.augmenters.blend.BlendAlphaMask`. Parameters ---------- factor : number or tuple of number or list of number or imgaug.parameters.StochasticParameter, optional Opacity of the results of the foreground branch. Values close to ``0.0`` mean that the results from the background branch (see parameter `background`) make up most of the final image. * If float, then that value will be used for all images. * If tuple ``(a, b)``, then a random value from the interval ``[a, b]`` will be sampled per image. * If a list, then a random value will be picked from that list per image. * If ``StochasticParameter``, then that parameter will be used to sample a value per image. foreground : None or imgaug.augmenters.meta.Augmenter or iterable of imgaug.augmenters.meta.Augmenter, optional Augmenter(s) that make up the foreground branch. High alpha values will show this branch's results. * If ``None``, then the input images will be reused as the output of the foreground branch. * If ``Augmenter``, then that augmenter will be used as the branch. * If iterable of ``Augmenter``, then that iterable will be converted into a ``Sequential`` and used as the augmenter. background : None or imgaug.augmenters.meta.Augmenter or iterable of imgaug.augmenters.meta.Augmenter, optional Augmenter(s) that make up the background branch. Low alpha values will show this branch's results. * If ``None``, then the input images will be reused as the output of the background branch. * If ``Augmenter``, then that augmenter will be used as the branch. * If iterable of ``Augmenter``, then that iterable will be converted into a ``Sequential`` and used as the augmenter. per_channel : bool or float, optional Whether to use the same factor for all channels (``False``) or to sample a new value for each channel (``True``). If this value is a float ``p``, then for ``p`` percent of all images `per_channel` will be treated as True, otherwise as False. seed : None or int or imgaug.random.RNG or numpy.random.Generator or numpy.random.BitGenerator or numpy.random.SeedSequence or numpy.random.RandomState, optional See :func:`~imgaug.augmenters.meta.Augmenter.__init__`. name : None or str, optional See :func:`~imgaug.augmenters.meta.Augmenter.__init__`. random_state : None or int or imgaug.random.RNG or numpy.random.Generator or numpy.random.BitGenerator or numpy.random.SeedSequence or numpy.random.RandomState, optional Old name for parameter `seed`. Its usage will not yet cause a deprecation warning, but it is still recommended to use `seed` now. Outdated since 0.4.0. deterministic : bool, optional Deprecated since 0.4.0. See method ``to_deterministic()`` for an alternative and for details about what the "deterministic mode" actually does. Examples -------- >>> import imgaug.augmenters as iaa >>> aug = iaa.BlendAlphaElementwise(0.5, iaa.Grayscale(1.0)) Convert each image to pure grayscale and alpha-blend the result with the original image using an alpha of ``50%`` for all pixels, thereby removing about ``50%`` of all color. This is equivalent to ``iaa.Grayscale(0.5)``. This is also equivalent to ``iaa.BlendAlpha(0.5, iaa.Grayscale(1.0))``, as the opacity has a fixed value of ``0.5`` and is hence identical for all pixels. >>> aug = iaa.BlendAlphaElementwise((0, 1.0), iaa.AddToHue(100)) Same as in the previous example, but here with hue-shift instead of grayscaling and additionally the alpha factor is sampled uniformly from the interval ``[0.0, 1.0]`` once per pixel, thereby shifting the hue by a random fraction for each pixel. >>> aug = iaa.BlendAlphaElementwise( >>> (0.0, 1.0), >>> iaa.Affine(rotate=(-20, 20)), >>> per_channel=0.5) First, rotate each image by a random degree sampled uniformly from the interval ``[-20, 20]``. Then, alpha-blend that new image with the original one using a random factor sampled uniformly from the interval ``[0.0, 1.0]`` per pixel. For ``50%`` of all images, the blending happens channel-wise and the factor is sampled independently per pixel *and* channel (``per_channel=0.5``). As a result, e.g. the red channel may look visibly rotated (factor near ``1.0``), while the green and blue channels may not look rotated (factors near ``0.0``). >>> aug = iaa.BlendAlphaElementwise( >>> (0.0, 1.0), >>> foreground=iaa.Add(100), >>> background=iaa.Multiply(0.2)) Apply two branches of augmenters -- ``A`` and ``B`` -- *independently* to input images and alpha-blend the results of these branches using a factor ``f``. Branch ``A`` increases image pixel intensities by ``100`` and ``B`` multiplies the pixel intensities by ``0.2``. ``f`` is sampled uniformly from the interval ``[0.0, 1.0]`` per pixel. The resulting images contain a bit of ``A`` and a bit of ``B``. >>> aug = iaa.BlendAlphaElementwise([0.25, 0.75], iaa.MedianBlur(13)) Apply median blur to each image and alpha-blend the result with the original image using an alpha factor of either exactly ``0.25`` or exactly ``0.75`` (sampled once per pixel). rGNFrIc s@tj|ddddd}t||} tt|j| ||||||ddS)NrLrMTrNrJ)rXrYStochasticParameterMaskGenrVrrW) r]rLr<r>rUrKr&rCrBZmask_genr^rrrWs   zBlendAlphaElementwise.__init__cC|jjSrw)r parameterryrrrrLrzBlendAlphaElementwise.factorr)rrrrrWpropertyrLrrrr^rr1s rcs.eZdZdZ      d fd d ZZS) BlendAlphaSimplexNoiseap#Alpha-blend two image sources using simplex noise alpha masks. The alpha masks are sampled using a simplex noise method, roughly creating connected blobs of 1s surrounded by 0s. If nearest neighbour upsampling is used, these blobs can be rectangular with sharp edges. Added in 0.4.0. (Before that named `SimplexNoiseAlpha`.) **Supported dtypes**: See :class:`~imgaug.augmenters.blend.BlendAlphaElementwise`. Parameters ---------- foreground : None or imgaug.augmenters.meta.Augmenter or iterable of imgaug.augmenters.meta.Augmenter, optional Augmenter(s) that make up the foreground branch. High alpha values will show this branch's results. * If ``None``, then the input images will be reused as the output of the foreground branch. * If ``Augmenter``, then that augmenter will be used as the branch. * If iterable of ``Augmenter``, then that iterable will be converted into a ``Sequential`` and used as the augmenter. background : None or imgaug.augmenters.meta.Augmenter or iterable of imgaug.augmenters.meta.Augmenter, optional Augmenter(s) that make up the background branch. Low alpha values will show this branch's results. * If ``None``, then the input images will be reused as the output of the background branch. * If ``Augmenter``, then that augmenter will be used as the branch. * If iterable of ``Augmenter``, then that iterable will be converted into a ``Sequential`` and used as the augmenter. per_channel : bool or float, optional Whether to use the same factor for all channels (``False``) or to sample a new value for each channel (``True``). If this value is a float ``p``, then for ``p`` percent of all images `per_channel` will be treated as ``True``, otherwise as ``False``. size_px_max : int or tuple of int or list of int or imgaug.parameters.StochasticParameter, optional The simplex noise is always generated in a low resolution environment. This parameter defines the maximum size of that environment (in pixels). The environment is initialized at the same size as the input image and then downscaled, so that no side exceeds `size_px_max` (aspect ratio is kept). * If int, then that number will be used as the size for all iterations. * If tuple of two ``int`` s ``(a, b)``, then a value will be sampled per iteration from the discrete interval ``[a..b]``. * If a list of ``int`` s, then a value will be picked per iteration at random from that list. * If a ``StochasticParameter``, then a value will be sampled from that parameter per iteration. upscale_method : None or imgaug.ALL or str or list of str or imgaug.parameters.StochasticParameter, optional After generating the noise maps in low resolution environments, they have to be upscaled to the input image size. This parameter controls the upscaling method. * If ``None``, then either ``nearest`` or ``linear`` or ``cubic`` is picked. Most weight is put on ``linear``, followed by ``cubic``. * If ``imgaug.ALL``, then either ``nearest`` or ``linear`` or ``area`` or ``cubic`` is picked per iteration (all same probability). * If a string, then that value will be used as the method (must be ``nearest`` or ``linear`` or ``area`` or ``cubic``). * If list of string, then a random value will be picked from that list per iteration. * If ``StochasticParameter``, then a random value will be sampled from that parameter per iteration. iterations : int or tuple of int or list of int or imgaug.parameters.StochasticParameter, optional How often to repeat the simplex noise generation process per image. * If ``int``, then that number will be used as the iterations for all images. * If tuple of two ``int`` s ``(a, b)``, then a value will be sampled per image from the discrete interval ``[a..b]``. * If a list of ``int`` s, then a value will be picked per image at random from that list. * If a ``StochasticParameter``, then a value will be sampled from that parameter per image. aggregation_method : imgaug.ALL or str or list of str or imgaug.parameters.StochasticParameter, optional The noise maps (from each iteration) are combined to one noise map using an aggregation process. This parameter defines the method used for that process. Valid methods are ``min``, ``max`` or ``avg``, where ``min`` combines the noise maps by taking the (elementwise) minimum over all iteration's results, ``max`` the (elementwise) maximum and ``avg`` the (elementwise) average. * If ``imgaug.ALL``, then a random value will be picked per image from the valid ones. * If a string, then that value will always be used as the method. * If a list of string, then a random value will be picked from that list per image. * If a ``StochasticParameter``, then a random value will be sampled from that paramter per image. sigmoid : bool or number, optional Whether to apply a sigmoid function to the final noise maps, resulting in maps that have more extreme values (close to 0.0 or 1.0). * If ``bool``, then a sigmoid will always (``True``) or never (``False``) be applied. * If a number ``p`` with ``0<=p<=1``, then a sigmoid will be applied to ``p`` percent of all final noise maps. sigmoid_thresh : None or number or tuple of number or imgaug.parameters.StochasticParameter, optional Threshold of the sigmoid, when applied. Thresholds above zero (e.g. ``5.0``) will move the saddle point towards the right, leading to more values close to 0.0. * If ``None``, then ``Normal(0, 5.0)`` will be used. * If number, then that threshold will be used for all images. * If tuple of two numbers ``(a, b)``, then a random value will be sampled per image from the interval ``[a, b]``. * If ``StochasticParameter``, then a random value will be sampled from that parameter per image. seed : None or int or imgaug.random.RNG or numpy.random.Generator or numpy.random.BitGenerator or numpy.random.SeedSequence or numpy.random.RandomState, optional See :func:`~imgaug.augmenters.meta.Augmenter.__init__`. name : None or str, optional See :func:`~imgaug.augmenters.meta.Augmenter.__init__`. random_state : None or int or imgaug.random.RNG or numpy.random.Generator or numpy.random.BitGenerator or numpy.random.SeedSequence or numpy.random.RandomState, optional Old name for parameter `seed`. Its usage will not yet cause a deprecation warning, but it is still recommended to use `seed` now. Outdated since 0.4.0. deterministic : bool, optional Deprecated since 0.4.0. See method ``to_deterministic()`` for an alternative and for details about what the "deterministic mode" actually does. Examples -------- >>> import imgaug.augmenters as iaa >>> aug = iaa.BlendAlphaSimplexNoise(iaa.EdgeDetect(1.0)) Detect per image all edges, mark them in a black and white image and then alpha-blend the result with the original image using simplex noise masks. >>> aug = iaa.BlendAlphaSimplexNoise( >>> iaa.EdgeDetect(1.0), >>> upscale_method="nearest") Same as in the previous example, but using only nearest neighbour upscaling to scale the simplex noise masks to the final image sizes, i.e. no nearest linear upsampling is used. This leads to rectangles with sharp edges. >>> aug = iaa.BlendAlphaSimplexNoise( >>> iaa.EdgeDetect(1.0), >>> upscale_method="linear") Same as in the previous example, but using only linear upscaling to scale the simplex noise masks to the final image sizes, i.e. no nearest neighbour upsampling is used. This leads to rectangles with smooth edges. >>> aug = iaa.BlendAlphaSimplexNoise( >>> iaa.EdgeDetect(1.0), >>> sigmoid_thresh=iap.Normal(10.0, 5.0)) Same as in the first example, but using a threshold for the sigmoid function that is further to the right. This is more conservative, i.e. the generated noise masks will be mostly black (values around ``0.0``), which means that most of the original images (parameter/branch `background`) will be kept, rather than using the results of the augmentation (parameter/branch `foreground`). NFr rrr3TrIc stjgdgdd}tdd}tj||dur|n|d}|dkr*tj|||d}|d up6t|o6|d k}|rHtjj|| durC| n||d }t t |j ||||| | | | d dS) NnearestZlinearZcubicg?g333333?gffffff?prH@) size_px_maxupscale_methodr iterationsaggregation_methodTr thresholdZ activatedrLr<r>rUrKr&rCrB) rXChoiceNormalZ SimplexNoiseIterativeNoiseAggregatorris_single_numberSigmoidcreate_for_noiserVrrW)r]r<r>rUrrrrsigmoidsigmoid_threshrKr&rCrBupscale_method_defaultsigmoid_thresh_defaultnoise use_sigmoidr^rrrWsB    zBlendAlphaSimplexNoise.__init__ NNFrNrr3TNNNrIrIrrrrrWrrrr^rrs5rcsBeZdZdZdddddddddgd dddd d ffd d ZZS) BlendAlphaFrequencyNoiseaT(Alpha-blend two image sources using frequency noise masks. The alpha masks are sampled using frequency noise of varying scales, which can sometimes create large connected blobs of ``1`` s surrounded by ``0`` s and other times results in smaller patterns. If nearest neighbour upsampling is used, these blobs can be rectangular with sharp edges. Added in 0.4.0. (Before that named `FrequencyNoiseAlpha`.) **Supported dtypes**: See :class:`~imgaug.augmenters.blend.BlendAlphaElementwise`. Parameters ---------- exponent : number or tuple of number of list of number or imgaug.parameters.StochasticParameter, optional Exponent to use when scaling in the frequency domain. Sane values are in the range ``-4`` (large blobs) to ``4`` (small patterns). To generate cloud-like structures, use roughly ``-2``. * If number, then that number will be used as the exponent for all iterations. * If tuple of two numbers ``(a, b)``, then a value will be sampled per iteration from the interval ``[a, b]``. * If a list of numbers, then a value will be picked per iteration at random from that list. * If a ``StochasticParameter``, then a value will be sampled from that parameter per iteration. foreground : None or imgaug.augmenters.meta.Augmenter or iterable of imgaug.augmenters.meta.Augmenter, optional Augmenter(s) that make up the foreground branch. High alpha values will show this branch's results. * If ``None``, then the input images will be reused as the output of the foreground branch. * If ``Augmenter``, then that augmenter will be used as the branch. * If iterable of ``Augmenter``, then that iterable will be converted into a ``Sequential`` and used as the augmenter. background : None or imgaug.augmenters.meta.Augmenter or iterable of imgaug.augmenters.meta.Augmenter, optional Augmenter(s) that make up the background branch. Low alpha values will show this branch's results. * If ``None``, then the input images will be reused as the output of the background branch. * If ``Augmenter``, then that augmenter will be used as the branch. * If iterable of ``Augmenter``, then that iterable will be converted into a ``Sequential`` and used as the augmenter. per_channel : bool or float, optional Whether to use the same factor for all channels (``False``) or to sample a new value for each channel (``True``). If this value is a float ``p``, then for ``p`` percent of all images `per_channel` will be treated as ``True``, otherwise as ``False``. size_px_max : int or tuple of int or list of int or imgaug.parameters.StochasticParameter, optional The noise is generated in a low resolution environment. This parameter defines the maximum size of that environment (in pixels). The environment is initialized at the same size as the input image and then downscaled, so that no side exceeds `size_px_max` (aspect ratio is kept). * If ``int``, then that number will be used as the size for all iterations. * If tuple of two ``int`` s ``(a, b)``, then a value will be sampled per iteration from the discrete interval ``[a..b]``. * If a list of ``int`` s, then a value will be picked per iteration at random from that list. * If a ``StochasticParameter``, then a value will be sampled from that parameter per iteration. upscale_method : None or imgaug.ALL or str or list of str or imgaug.parameters.StochasticParameter, optional After generating the noise maps in low resolution environments, they have to be upscaled to the input image size. This parameter controls the upscaling method. * If ``None``, then either ``nearest`` or ``linear`` or ``cubic`` is picked. Most weight is put on ``linear``, followed by ``cubic``. * If ``imgaug.ALL``, then either ``nearest`` or ``linear`` or ``area`` or ``cubic`` is picked per iteration (all same probability). * If string, then that value will be used as the method (must be ``nearest`` or ``linear`` or ``area`` or ``cubic``). * If list of string, then a random value will be picked from that list per iteration. * If ``StochasticParameter``, then a random value will be sampled from that parameter per iteration. iterations : int or tuple of int or list of int or imgaug.parameters.StochasticParameter, optional How often to repeat the simplex noise generation process per image. * If ``int``, then that number will be used as the iterations for all images. * If tuple of two ``int`` s ``(a, b)``, then a value will be sampled per image from the discrete interval ``[a..b]``. * If a list of ``int`` s, then a value will be picked per image at random from that list. * If a ``StochasticParameter``, then a value will be sampled from that parameter per image. aggregation_method : imgaug.ALL or str or list of str or imgaug.parameters.StochasticParameter, optional The noise maps (from each iteration) are combined to one noise map using an aggregation process. This parameter defines the method used for that process. Valid methods are ``min``, ``max`` or ``avg``, where 'min' combines the noise maps by taking the (elementwise) minimum over all iteration's results, ``max`` the (elementwise) maximum and ``avg`` the (elementwise) average. * If ``imgaug.ALL``, then a random value will be picked per image from the valid ones. * If a string, then that value will always be used as the method. * If a list of string, then a random value will be picked from that list per image. * If a ``StochasticParameter``, then a random value will be sampled from that parameter per image. sigmoid : bool or number, optional Whether to apply a sigmoid function to the final noise maps, resulting in maps that have more extreme values (close to ``0.0`` or ``1.0``). * If ``bool``, then a sigmoid will always (``True``) or never (``False``) be applied. * If a number ``p`` with ``0<=p<=1``, then a sigmoid will be applied to ``p`` percent of all final noise maps. sigmoid_thresh : None or number or tuple of number or imgaug.parameters.StochasticParameter, optional Threshold of the sigmoid, when applied. Thresholds above zero (e.g. ``5.0``) will move the saddle point towards the right, leading to more values close to ``0.0``. * If ``None``, then ``Normal(0, 5.0)`` will be used. * If number, then that threshold will be used for all images. * If tuple of two numbers ``(a, b)``, then a random value will be sampled per image from the range ``[a, b]``. * If ``StochasticParameter``, then a random value will be sampled from that parameter per image. seed : None or int or imgaug.random.RNG or numpy.random.Generator or numpy.random.BitGenerator or numpy.random.SeedSequence or numpy.random.RandomState, optional See :func:`~imgaug.augmenters.meta.Augmenter.__init__`. name : None or str, optional See :func:`~imgaug.augmenters.meta.Augmenter.__init__`. random_state : None or int or imgaug.random.RNG or numpy.random.Generator or numpy.random.BitGenerator or numpy.random.SeedSequence or numpy.random.RandomState, optional Old name for parameter `seed`. Its usage will not yet cause a deprecation warning, but it is still recommended to use `seed` now. Outdated since 0.4.0. deterministic : bool, optional Deprecated since 0.4.0. See method ``to_deterministic()`` for an alternative and for details about what the "deterministic mode" actually does. Examples -------- >>> import imgaug.augmenters as iaa >>> aug = iaa.BlendAlphaFrequencyNoise(foreground=iaa.EdgeDetect(1.0)) Detect per image all edges, mark them in a black and white image and then alpha-blend the result with the original image using frequency noise masks. >>> aug = iaa.BlendAlphaFrequencyNoise( >>> foreground=iaa.EdgeDetect(1.0), >>> upscale_method="nearest") Same as the first example, but using only linear upscaling to scale the frequency noise masks to the final image sizes, i.e. no nearest neighbour upsampling is used. This results in smooth edges. >>> aug = iaa.BlendAlphaFrequencyNoise( >>> foreground=iaa.EdgeDetect(1.0), >>> upscale_method="linear") Same as the first example, but using only linear upscaling to scale the frequency noise masks to the final image sizes, i.e. no nearest neighbour upsampling is used. This results in smooth edges. >>> aug = iaa.BlendAlphaFrequencyNoise( >>> foreground=iaa.EdgeDetect(1.0), >>> upscale_method="linear", >>> exponent=-2, >>> sigmoid=False) Same as in the previous example, but with the exponent set to a constant ``-2`` and the sigmoid deactivated, resulting in cloud-like patterns without sharp edges. >>> aug = iaa.BlendAlphaFrequencyNoise( >>> foreground=iaa.EdgeDetect(1.0), >>> sigmoid_thresh=iap.Normal(10.0, 5.0)) Same as the first example, but using a threshold for the sigmoid function that is further to the right. This is more conservative, i.e. the generated noise masks will be mostly black (values around ``0.0``), which means that most of the original images (parameter/branch `background`) will be kept, rather than using the results of the augmentation (parameter/branch `foreground`). r NFr rravgr3r!rIc stjgdgdd}tdd}tj|||dur|n|d}|dkr+tj|||d}| d up7t| o7| d k}|rItjj|| durD| n|| d }t t |j ||||| | | |d dS) NrrrrHr)exponentrrrrTrrr) rXrrZFrequencyNoiserrrrrrVrrW)r]rr<r>rUrrrrrrrKr&rCrBrrrrr^rrrW|sD    z!BlendAlphaFrequencyNoise.__init__rrrr^rrsOrc s<eZdZdZddddddgddddd d f fd d ZZS) BlendAlphaSomeColorsaBlend images from two branches using colorwise masks. This class generates masks that "mark" a few colors and replace the pixels within these colors with the results of the foreground branch. The remaining pixels are replaced with the results of the background branch (usually the identity function). That allows to e.g. selectively grayscale a few colors, while keeping other colors unchanged. This class is a thin wrapper around :class:`~imgaug.augmenters.blend.BlendAlphaMask` together with :class:`~imgaug.augmenters.blend.SomeColorsMaskGen`. .. note:: The underlying mask generator will produce an ``AssertionError`` for batches that contain no images. .. note:: Avoid using augmenters as children that affect pixel locations (e.g. horizontal flips). See :class:`~imgaug.augmenters.blend.BlendAlphaMask` for details. Added in 0.4.0. **Supported dtypes**: See :func:`~imgaug.augmenters.color.change_colorspaces_`. Parameters ---------- foreground : None or imgaug.augmenters.meta.Augmenter or iterable of imgaug.augmenters.meta.Augmenter, optional Augmenter(s) that make up the foreground branch. High alpha values will show this branch's results. * If ``None``, then the input images will be reused as the output of the foreground branch. * If ``Augmenter``, then that augmenter will be used as the branch. * If iterable of ``Augmenter``, then that iterable will be converted into a ``Sequential`` and used as the augmenter. background : None or imgaug.augmenters.meta.Augmenter or iterable of imgaug.augmenters.meta.Augmenter, optional Augmenter(s) that make up the background branch. Low alpha values will show this branch's results. * If ``None``, then the input images will be reused as the output of the background branch. * If ``Augmenter``, then that augmenter will be used as the branch. * If iterable of ``Augmenter``, then that iterable will be converted into a ``Sequential`` and used as the augmenter. nb_bins : int or tuple of int or list of int or imgaug.parameters.StochasticParameter, optional See :class:`~imgaug.augmenters.blend.SomeColorsMaskGen`. smoothness : number or tuple of number or list of number or imgaug.parameters.StochasticParameter, optional See :class:`~imgaug.augmenters.blend.SomeColorsMaskGen`. alpha : number or tuple of number or list of number or imgaug.parameters.StochasticParameter, optional See :class:`~imgaug.augmenters.blend.SomeColorsMaskGen`. rotation_deg : number or tuple of number or list of number or imgaug.parameters.StochasticParameter, optional See :class:`~imgaug.augmenters.blend.SomeColorsMaskGen`. from_colorspace : str, optional See :class:`~imgaug.augmenters.blend.SomeColorsMaskGen`. seed : None or int or imgaug.random.RNG or numpy.random.Generator or numpy.random.BitGenerator or numpy.random.SeedSequence or numpy.random.RandomState, optional See :func:`~imgaug.augmenters.meta.Augmenter.__init__`. name : None or str, optional See :func:`~imgaug.augmenters.meta.Augmenter.__init__`. random_state : None or int or imgaug.random.RNG or numpy.random.Generator or numpy.random.BitGenerator or numpy.random.SeedSequence or numpy.random.RandomState, optional Old name for parameter `seed`. Its usage will not yet cause a deprecation warning, but it is still recommended to use `seed` now. Outdated since 0.4.0. deterministic : bool, optional Deprecated since 0.4.0. See method ``to_deterministic()`` for an alternative and for details about what the "deterministic mode" actually does. Examples -------- >>> import imgaug.augmenters as iaa >>> aug = iaa.BlendAlphaSomeColors(iaa.Grayscale(1.0)) Create an augmenter that turns randomly removes some colors in images by grayscaling them. >>> aug = iaa.BlendAlphaSomeColors(iaa.TotalDropout(1.0)) Create an augmenter that removes some colors in images by replacing them with black pixels. >>> aug = iaa.BlendAlphaSomeColors( >>> iaa.MultiplySaturation(0.5), iaa.MultiplySaturation(1.5)) Create an augmenter that desaturates some colors and increases the saturation of the remaining ones. >>> aug = iaa.BlendAlphaSomeColors( >>> iaa.AveragePooling(7), alpha=[0.0, 1.0], smoothness=0.0) Create an augmenter that applies average pooling to some colors. Each color tune is either selected (alpha of ``1.0``) or not selected (``0.0``). There is no gradual change between similar colors. >>> aug = iaa.BlendAlphaSomeColors( >>> iaa.AveragePooling(7), nb_bins=2, smoothness=0.0) Create an augmenter that applies average pooling to some colors. Choose on average half of all colors in images for the blending operation. >>> aug = iaa.BlendAlphaSomeColors( >>> iaa.AveragePooling(7), from_colorspace="BGR") Create an augmenter that applies average pooling to some colors with input images being in BGR colorspace. Ng?g333333?rHrrhRGBrIc s0tt|jt|||||d|||| | | ddS)N)nb_bins smoothnessr6 rotation_degfrom_colorspacer<r>rKr&rCrB)rVrrWSomeColorsMaskGen) r]r<r>rrr6rrrKr&rCrBr^rrrW&s  zBlendAlphaSomeColors.__init__rrrr^rrs|rc,eZdZdZ     dfdd ZZS) "BlendAlphaHorizontalLinearGradientaKBlend images from two branches along a horizontal linear gradient. This class generates a horizontal linear gradient mask (i.e. usually a mask with low values on the left and high values on the right) and alphas-blends between foreground and background branch using that mask. This class is a thin wrapper around :class:`~imgaug.augmenters.blend.BlendAlphaMask` together with :class:`~imgaug.augmenters.blend.HorizontalLinearGradientMaskGen`. .. note:: Avoid using augmenters as children that affect pixel locations (e.g. horizontal flips). See :class:`~imgaug.augmenters.blend.BlendAlphaMask` for details. Added in 0.4.0. **Supported dtypes**: See :class:`~imgaug.augmenters.blend.BlendAlphaMask`. Parameters ---------- foreground : None or imgaug.augmenters.meta.Augmenter or iterable of imgaug.augmenters.meta.Augmenter, optional Augmenter(s) that make up the foreground branch. High alpha values will show this branch's results. * If ``None``, then the input images will be reused as the output of the foreground branch. * If ``Augmenter``, then that augmenter will be used as the branch. * If iterable of ``Augmenter``, then that iterable will be converted into a ``Sequential`` and used as the augmenter. background : None or imgaug.augmenters.meta.Augmenter or iterable of imgaug.augmenters.meta.Augmenter, optional Augmenter(s) that make up the background branch. Low alpha values will show this branch's results. * If ``None``, then the input images will be reused as the output of the background branch. * If ``Augmenter``, then that augmenter will be used as the branch. * If iterable of ``Augmenter``, then that iterable will be converted into a ``Sequential`` and used as the augmenter. min_value : number or tuple of number or list of number or imgaug.parameters.StochasticParameter, optional See :class:`~imgaug.augmenters.blend.HorizontalLinearGradientMaskGen`. max_value : number or tuple of number or list of number or imgaug.parameters.StochasticParameter, optional See :class:`~imgaug.augmenters.blend.HorizontalLinearGradientMaskGen`. start_at : number or tuple of number or list of number or imgaug.parameters.StochasticParameter, optional See :class:`~imgaug.augmenters.blend.HorizontalLinearGradientMaskGen`. end_at : number or tuple of number or list of number or imgaug.parameters.StochasticParameter, optional See :class:`~imgaug.augmenters.blend.HorizontalLinearGradientMaskGen`. seed : None or int or imgaug.random.RNG or numpy.random.Generator or numpy.random.BitGenerator or numpy.random.SeedSequence or numpy.random.RandomState, optional See :func:`~imgaug.augmenters.meta.Augmenter.__init__`. name : None or str, optional See :func:`~imgaug.augmenters.meta.Augmenter.__init__`. random_state : None or int or imgaug.random.RNG or numpy.random.Generator or numpy.random.BitGenerator or numpy.random.SeedSequence or numpy.random.RandomState, optional Old name for parameter `seed`. Its usage will not yet cause a deprecation warning, but it is still recommended to use `seed` now. Outdated since 0.4.0. deterministic : bool, optional Deprecated since 0.4.0. See method ``to_deterministic()`` for an alternative and for details about what the "deterministic mode" actually does. Examples -------- >>> import imgaug.augmenters as iaa >>> aug = iaa.BlendAlphaHorizontalLinearGradient(iaa.AddToHue((-100, 100))) Create an augmenter that randomizes the hue towards the right of the image. >>> aug = iaa.BlendAlphaHorizontalLinearGradient( >>> iaa.TotalDropout(1.0), >>> min_value=0.2, max_value=0.8) Create an augmenter that replaces pixels towards the right with darker and darker values. However it always keeps at least 20% (``1.0 - max_value``) of the original pixel value on the far right and always replaces at least 20% on the far left (``min_value=0.2``). >>> aug = iaa.BlendAlphaHorizontalLinearGradient( >>> iaa.AveragePooling(11), >>> start_at=(0.0, 1.0), end_at=(0.0, 1.0)) Create an augmenter that blends with an average-pooled image according to a horizontal gradient that starts at a random x-coordinate and reaches its maximum at another random x-coordinate. Due to that randomness, the gradient may increase towards the left or right. NrHg?g?rrIc .tt|jt||||d||||| | ddSN) min_value max_valuestart_atend_atr)rVrrWHorizontalLinearGradientMaskGen r]r<r>rrrrrKr&rCrBr^rrrW  z+BlendAlphaHorizontalLinearGradient.__init__ NNrrrrNNrIrIrrrr^rr;sgrcr) BlendAlphaVerticalLinearGradienta Blend images from two branches along a vertical linear gradient. This class generates a vertical linear gradient mask (i.e. usually a mask with low values on the left and high values on the right) and alphas-blends between foreground and background branch using that mask. This class is a thin wrapper around :class:`~imgaug.augmenters.blend.BlendAlphaMask` together with :class:`~imgaug.augmenters.blend.VerticalLinearGradientMaskGen`. .. note:: Avoid using augmenters as children that affect pixel locations (e.g. horizontal flips). See :class:`~imgaug.augmenters.blend.BlendAlphaMask` for details. Added in 0.4.0. **Supported dtypes**: See :class:`~imgaug.augmenters.blend.BlendAlphaMask`. Parameters ---------- foreground : None or imgaug.augmenters.meta.Augmenter or iterable of imgaug.augmenters.meta.Augmenter, optional Augmenter(s) that make up the foreground branch. High alpha values will show this branch's results. * If ``None``, then the input images will be reused as the output of the foreground branch. * If ``Augmenter``, then that augmenter will be used as the branch. * If iterable of ``Augmenter``, then that iterable will be converted into a ``Sequential`` and used as the augmenter. background : None or imgaug.augmenters.meta.Augmenter or iterable of imgaug.augmenters.meta.Augmenter, optional Augmenter(s) that make up the background branch. Low alpha values will show this branch's results. * If ``None``, then the input images will be reused as the output of the background branch. * If ``Augmenter``, then that augmenter will be used as the branch. * If iterable of ``Augmenter``, then that iterable will be converted into a ``Sequential`` and used as the augmenter. min_value : number or tuple of number or list of number or imgaug.parameters.StochasticParameter, optional See :class:`~imgaug.augmenters.blend.VerticalLinearGradientMaskGen`. max_value : number or tuple of number or list of number or imgaug.parameters.StochasticParameter, optional See :class:`~imgaug.augmenters.blend.VerticalLinearGradientMaskGen`. start_at : number or tuple of number or list of number or imgaug.parameters.StochasticParameter, optional See :class:`~imgaug.augmenters.blend.VerticalLinearGradientMaskGen`. end_at : number or tuple of number or list of number or imgaug.parameters.StochasticParameter, optional See :class:`~imgaug.augmenters.blend.VerticalLinearGradientMaskGen`. seed : None or int or imgaug.random.RNG or numpy.random.Generator or numpy.random.BitGenerator or numpy.random.SeedSequence or numpy.random.RandomState, optional See :func:`~imgaug.augmenters.meta.Augmenter.__init__`. name : None or str, optional See :func:`~imgaug.augmenters.meta.Augmenter.__init__`. random_state : None or int or imgaug.random.RNG or numpy.random.Generator or numpy.random.BitGenerator or numpy.random.SeedSequence or numpy.random.RandomState, optional Old name for parameter `seed`. Its usage will not yet cause a deprecation warning, but it is still recommended to use `seed` now. Outdated since 0.4.0. deterministic : bool, optional Deprecated since 0.4.0. See method ``to_deterministic()`` for an alternative and for details about what the "deterministic mode" actually does. Examples -------- >>> import imgaug.augmenters as iaa >>> aug = iaa.BlendAlphaVerticalLinearGradient(iaa.AddToHue((-100, 100))) Create an augmenter that randomizes the hue towards the bottom of the image. >>> aug = iaa.BlendAlphaVerticalLinearGradient( >>> iaa.TotalDropout(1.0), >>> min_value=0.2, max_value=0.8) Create an augmenter that replaces pixels towards the bottom with darker and darker values. However it always keeps at least 20% (``1.0 - max_value``) of the original pixel value on the far bottom and always replaces at least 20% on the far top (``min_value=0.2``). >>> aug = iaa.BlendAlphaVerticalLinearGradient( >>> iaa.AveragePooling(11), >>> start_at=(0.0, 1.0), end_at=(0.0, 1.0)) Create an augmenter that blends with an average-pooled image according to a vertical gradient that starts at a random y-coordinate and reaches its maximum at another random y-coordinate. Due to that randomness, the gradient may increase towards the bottom or top. >>> aug = iaa.BlendAlphaVerticalLinearGradient( >>> iaa.Clouds(), >>> start_at=(0.15, 0.35), end_at=0.0) Create an augmenter that draws clouds in roughly the top quarter of the image. NrrrIc rr)rVrrWVerticalLinearGradientMaskGenrr^rrrW$rz)BlendAlphaVerticalLinearGradient.__init__rrrrr^rrsnrcs4eZdZdZddddgddddffdd ZZS)BlendAlphaRegularGridaEBlend images from two branches according to a regular grid. This class generates for each image a mask that splits the image into a grid-like pattern of ``H`` rows and ``W`` columns. Each cell is then filled with an alpha value, sampled randomly per cell. The difference to :class:`AlphaBlendCheckerboard` is that this class samples random alpha values per grid cell, while in the checkerboard the alpha values follow a fixed pattern. This class is a thin wrapper around :class:`~imgaug.augmenters.blend.BlendAlphaMask` together with :class:`~imgaug.augmenters.blend.RegularGridMaskGen`. .. note:: Avoid using augmenters as children that affect pixel locations (e.g. horizontal flips). See :class:`~imgaug.augmenters.blend.BlendAlphaMask` for details. Added in 0.4.0. **Supported dtypes**: See :class:`~imgaug.augmenters.blend.BlendAlphaMask`. Parameters ---------- nb_rows : int or tuple of int or list of int or imgaug.parameters.StochasticParameter Number of rows of the checkerboard. See :class:`~imgaug.augmenters.blend.CheckerboardMaskGen` for details. nb_cols : int or tuple of int or list of int or imgaug.parameters.StochasticParameter Number of columns of the checkerboard. Analogous to `nb_rows`. See :class:`~imgaug.augmenters.blend.CheckerboardMaskGen` for details. foreground : None or imgaug.augmenters.meta.Augmenter or iterable of imgaug.augmenters.meta.Augmenter, optional Augmenter(s) that make up the foreground branch. High alpha values will show this branch's results. * If ``None``, then the input images will be reused as the output of the foreground branch. * If ``Augmenter``, then that augmenter will be used as the branch. * If iterable of ``Augmenter``, then that iterable will be converted into a ``Sequential`` and used as the augmenter. background : None or imgaug.augmenters.meta.Augmenter or iterable of imgaug.augmenters.meta.Augmenter, optional Augmenter(s) that make up the background branch. Low alpha values will show this branch's results. * If ``None``, then the input images will be reused as the output of the background branch. * If ``Augmenter``, then that augmenter will be used as the branch. * If iterable of ``Augmenter``, then that iterable will be converted into a ``Sequential`` and used as the augmenter. alpha : number or tuple of number or list of number or imgaug.parameters.StochasticParameter, optional Alpha value of each cell. * If ``number``: Exactly that value will be used for all images. * If ``tuple`` ``(a, b)``: A random value will be uniformly sampled per image from the interval ``[a, b]``. * If ``list``: A random value will be picked per image from that list. * If ``StochasticParameter``: That parameter will be queried once per batch for ``(N,)`` values -- one per image. seed : None or int or imgaug.random.RNG or numpy.random.Generator or numpy.random.BitGenerator or numpy.random.SeedSequence or numpy.random.RandomState, optional See :func:`~imgaug.augmenters.meta.Augmenter.__init__`. name : None or str, optional See :func:`~imgaug.augmenters.meta.Augmenter.__init__`. random_state : None or int or imgaug.random.RNG or numpy.random.Generator or numpy.random.BitGenerator or numpy.random.SeedSequence or numpy.random.RandomState, optional Old name for parameter `seed`. Its usage will not yet cause a deprecation warning, but it is still recommended to use `seed` now. Outdated since 0.4.0. deterministic : bool, optional Deprecated since 0.4.0. See method ``to_deterministic()`` for an alternative and for details about what the "deterministic mode" actually does. Examples -------- >>> import imgaug.augmenters as iaa >>> aug = iaa.BlendAlphaRegularGrid(nb_rows=(4, 6), nb_cols=(1, 4), >>> foreground=iaa.Multiply(0.0)) Create an augmenter that places a ``HxW`` grid on each image, where ``H`` (rows) is randomly and uniformly sampled from the interval ``[4, 6]`` and ``W`` is analogously sampled from the interval ``[1, 4]``. Roughly half of the cells in the grid are filled with ``0.0``, the remaining ones are unaltered. Which cells exactly are "dropped" is randomly decided per image. The resulting effect is similar to :class:`~imgaug.augmenters.arithmetic.CoarseDropout`. >>> aug = iaa.BlendAlphaRegularGrid(nb_rows=2, nb_cols=2, >>> foreground=iaa.Multiply(0.0), >>> background=iaa.AveragePooling(8), >>> alpha=[0.0, 0.0, 1.0]) Create an augmenter that always placed ``2x2`` cells on each image and sets about ``1/3`` of them to zero (foreground branch) and the remaining ``2/3`` to a pixelated version (background branch). NrHrrIc s,tt|jt|||d|||||| ddS)Nnb_rowsnb_colsr6r)rVrrWRegularGridMaskGen) r]rrr<r>r6rKr&rCrBr^rrrWs  zBlendAlphaRegularGrid.__init__rrrr^rr6snrcs(eZdZdZ   dfdd ZZS)BlendAlphaCheckerboardatBlend images from two branches according to a checkerboard pattern. This class generates for each image a mask following a checkboard layout of ``H`` rows and ``W`` columns. Each cell is then filled with either ``1.0`` or ``0.0``. The cell at the top-left is always ``1.0``. Its right and bottom neighbour cells are ``0.0``. The 4-neighbours of any cell always have a value opposite to the cell's value (``0.0`` vs. ``1.0``). This class is a thin wrapper around :class:`~imgaug.augmenters.blend.BlendAlphaMask` together with :class:`~imgaug.augmenters.blend.CheckerboardMaskGen`. .. note:: Avoid using augmenters as children that affect pixel locations (e.g. horizontal flips). See :class:`~imgaug.augmenters.blend.BlendAlphaMask` for details. Added in 0.4.0. **Supported dtypes**: See :class:`~imgaug.augmenters.blend.BlendAlphaMask`. Parameters ---------- nb_rows : int or tuple of int or list of int or imgaug.parameters.StochasticParameter Number of rows of the checkerboard. See :class:`~imgaug.augmenters.blend.CheckerboardMaskGen` for details. nb_cols : int or tuple of int or list of int or imgaug.parameters.StochasticParameter Number of columns of the checkerboard. Analogous to `nb_rows`. See :class:`~imgaug.augmenters.blend.CheckerboardMaskGen` for details. foreground : None or imgaug.augmenters.meta.Augmenter or iterable of imgaug.augmenters.meta.Augmenter, optional Augmenter(s) that make up the foreground branch. High alpha values will show this branch's results. * If ``None``, then the input images will be reused as the output of the foreground branch. * If ``Augmenter``, then that augmenter will be used as the branch. * If iterable of ``Augmenter``, then that iterable will be converted into a ``Sequential`` and used as the augmenter. background : None or imgaug.augmenters.meta.Augmenter or iterable of imgaug.augmenters.meta.Augmenter, optional Augmenter(s) that make up the background branch. Low alpha values will show this branch's results. * If ``None``, then the input images will be reused as the output of the background branch. * If ``Augmenter``, then that augmenter will be used as the branch. * If iterable of ``Augmenter``, then that iterable will be converted into a ``Sequential`` and used as the augmenter. seed : None or int or imgaug.random.RNG or numpy.random.Generator or numpy.random.BitGenerator or numpy.random.SeedSequence or numpy.random.RandomState, optional See :func:`~imgaug.augmenters.meta.Augmenter.__init__`. name : None or str, optional See :func:`~imgaug.augmenters.meta.Augmenter.__init__`. random_state : None or int or imgaug.random.RNG or numpy.random.Generator or numpy.random.BitGenerator or numpy.random.SeedSequence or numpy.random.RandomState, optional Old name for parameter `seed`. Its usage will not yet cause a deprecation warning, but it is still recommended to use `seed` now. Outdated since 0.4.0. deterministic : bool, optional Deprecated since 0.4.0. See method ``to_deterministic()`` for an alternative and for details about what the "deterministic mode" actually does. Examples -------- >>> import imgaug.augmenters as iaa >>> aug = iaa.BlendAlphaCheckerboard(nb_rows=2, nb_cols=(1, 4), >>> foreground=iaa.AddToHue((-100, 100))) Create an augmenter that places a ``HxW`` grid on each image, where ``H`` (rows) is always ``2`` and ``W`` is randomly and uniformly sampled from the interval ``[1, 4]``. For half of the cells in the grid the hue is randomly modified, the other half of the cells is unaltered. NrIc s*tt|jt||d||||||ddS)N)rrr)rVrrWCheckerboardMaskGen) r]rrr<r>rKr&rCrBr^rrrW s  zBlendAlphaCheckerboard.__init__rrrrr^rrs Vrc*eZdZdZ    dfdd ZZS)BlendAlphaSegMapClassIdsaBlend images from two branches based on segmentation map ids. This class generates masks that are ``1.0`` at pixel locations covered by specific classes in segmentation maps. This class is a thin wrapper around :class:`~imgaug.augmenters.blend.BlendAlphaMask` together with :class:`~imgaug.augmenters.blend.SegMapClassIdsMaskGen`. .. note:: Avoid using augmenters as children that affect pixel locations (e.g. horizontal flips). See :class:`~imgaug.augmenters.blend.BlendAlphaMask` for details. .. note:: Segmentation maps can have multiple channels. If that is the case then for each position ``(x, y)`` it is sufficient that any class id in any channel matches one of the desired class ids. .. note:: This class will produce an ``AssertionError`` if there are no segmentation maps in a batch. Added in 0.4.0. **Supported dtypes**: See :class:`~imgaug.augmenters.blend.BlendAlphaMask`. Parameters ---------- class_ids : int or tuple of int or list of int or imgaug.parameters.StochasticParameter See :class:`~imgaug.augmenters.blend.SegMapClassIdsMaskGen`. foreground : None or imgaug.augmenters.meta.Augmenter or iterable of imgaug.augmenters.meta.Augmenter, optional Augmenter(s) that make up the foreground branch. High alpha values will show this branch's results. * If ``None``, then the input images will be reused as the output of the foreground branch. * If ``Augmenter``, then that augmenter will be used as the branch. * If iterable of ``Augmenter``, then that iterable will be converted into a ``Sequential`` and used as the augmenter. background : None or imgaug.augmenters.meta.Augmenter or iterable of imgaug.augmenters.meta.Augmenter, optional Augmenter(s) that make up the background branch. Low alpha values will show this branch's results. * If ``None``, then the input images will be reused as the output of the background branch. * If ``Augmenter``, then that augmenter will be used as the branch. * If iterable of ``Augmenter``, then that iterable will be converted into a ``Sequential`` and used as the augmenter. nb_sample_classes : None or tuple of int or list of int or imgaug.parameters.StochasticParameter, optional See :class:`~imgaug.augmenters.blend.SegMapClassIdsMaskGen`. seed : None or int or imgaug.random.RNG or numpy.random.Generator or numpy.random.BitGenerator or numpy.random.SeedSequence or numpy.random.RandomState, optional See :func:`~imgaug.augmenters.meta.Augmenter.__init__`. name : None or str, optional See :func:`~imgaug.augmenters.meta.Augmenter.__init__`. random_state : None or int or imgaug.random.RNG or numpy.random.Generator or numpy.random.BitGenerator or numpy.random.SeedSequence or numpy.random.RandomState, optional Old name for parameter `seed`. Its usage will not yet cause a deprecation warning, but it is still recommended to use `seed` now. Outdated since 0.4.0. deterministic : bool, optional Deprecated since 0.4.0. See method ``to_deterministic()`` for an alternative and for details about what the "deterministic mode" actually does. Examples -------- >>> import imgaug.augmenters as iaa >>> aug = iaa.BlendAlphaSegMapClassIds( >>> [1, 3], >>> foreground=iaa.AddToHue((-100, 100))) Create an augmenter that randomizes the hue wherever the segmentation maps contain the classes ``1`` or ``3``. >>> aug = iaa.BlendAlphaSegMapClassIds( >>> [1, 2, 3, 4], >>> nb_sample_classes=2, >>> foreground=iaa.GaussianBlur(3.0)) Create an augmenter that randomly picks ``2`` classes from the list ``[1, 2, 3, 4]`` and blurs the image content wherever these classes appear in the segmentation map. Note that as the sampling of class ids happens *with replacement*, it is not guaranteed to sample two *unique* class ids. >>> aug = iaa.Sometimes(0.2, >>> iaa.BlendAlphaSegMapClassIds( >>> 2, >>> background=iaa.TotalDropout(1.0))) Create an augmenter that zeros for roughly every fifth image all image pixels that do *not* belong to class id ``2`` (note that the `background` branch was used, not the `foreground` branch). Example use case: Human body landmark detection where both the landmarks/keypoints and the body segmentation map are known. Train the model to detect landmarks and sometimes remove all non-body information to force the model to become more independent of the background. NrIc *tt|jt||d||||||ddS)N) class_idsnb_sample_classesr)rVrrWSegMapClassIdsMaskGen) r]rr<r>rrKr&rCrBr^rrrW  z!BlendAlphaSegMapClassIds.__init__NNNNNrIrIrrrr^rrstrcr)BlendAlphaBoundingBoxesa(Blend images from two branches based on areas enclosed in bounding boxes. This class generates masks that are ``1.0`` within bounding boxes of given labels. A mask pixel will be set to ``1.0`` if *at least* one bounding box covers the area and has one of the requested labels. This class is a thin wrapper around :class:`~imgaug.augmenters.blend.BlendAlphaMask` together with :class:`~imgaug.augmenters.blend.BoundingBoxesMaskGen`. .. note:: Avoid using augmenters as children that affect pixel locations (e.g. horizontal flips). See :class:`~imgaug.augmenters.blend.BlendAlphaMask` for details. .. note:: This class will produce an ``AssertionError`` if there are no bounding boxes in a batch. Added in 0.4.0. **Supported dtypes**: See :class:`~imgaug.augmenters.blend.BlendAlphaMask`. Parameters ---------- labels : None or str or list of str or imgaug.parameters.StochasticParameter See :class:`~imgaug.augmenters.blend.BoundingBoxesMaskGen`. foreground : None or imgaug.augmenters.meta.Augmenter or iterable of imgaug.augmenters.meta.Augmenter, optional Augmenter(s) that make up the foreground branch. High alpha values will show this branch's results. * If ``None``, then the input images will be reused as the output of the foreground branch. * If ``Augmenter``, then that augmenter will be used as the branch. * If iterable of ``Augmenter``, then that iterable will be converted into a ``Sequential`` and used as the augmenter. background : None or imgaug.augmenters.meta.Augmenter or iterable of imgaug.augmenters.meta.Augmenter, optional Augmenter(s) that make up the background branch. Low alpha values will show this branch's results. * If ``None``, then the input images will be reused as the output of the background branch. * If ``Augmenter``, then that augmenter will be used as the branch. * If iterable of ``Augmenter``, then that iterable will be converted into a ``Sequential`` and used as the augmenter. nb_sample_labels : None or tuple of int or list of int or imgaug.parameters.StochasticParameter, optional See :class:`~imgaug.augmenters.blend.BoundingBoxesMaskGen`. seed : None or int or imgaug.random.RNG or numpy.random.Generator or numpy.random.BitGenerator or numpy.random.SeedSequence or numpy.random.RandomState, optional See :func:`~imgaug.augmenters.meta.Augmenter.__init__`. name : None or str, optional See :func:`~imgaug.augmenters.meta.Augmenter.__init__`. random_state : None or int or imgaug.random.RNG or numpy.random.Generator or numpy.random.BitGenerator or numpy.random.SeedSequence or numpy.random.RandomState, optional Old name for parameter `seed`. Its usage will not yet cause a deprecation warning, but it is still recommended to use `seed` now. Outdated since 0.4.0. deterministic : bool, optional Deprecated since 0.4.0. See method ``to_deterministic()`` for an alternative and for details about what the "deterministic mode" actually does. Examples -------- >>> import imgaug.augmenters as iaa >>> aug = iaa.BlendAlphaBoundingBoxes("person", >>> foreground=iaa.Grayscale(1.0)) Create an augmenter that removes color within bounding boxes having the label ``person``. >>> aug = iaa.BlendAlphaBoundingBoxes(["person", "car"], >>> foreground=iaa.AddToHue((-255, 255))) Create an augmenter that randomizes the hue within bounding boxes that have the label ``person`` or ``car``. >>> aug = iaa.BlendAlphaBoundingBoxes(["person", "car"], >>> foreground=iaa.AddToHue((-255, 255)), >>> nb_sample_labels=1) Create an augmenter that randomizes the hue within bounding boxes that have either the label ``person`` or ``car``. Only one label is picked per image. Note that the sampling happens with replacement, so if ``nb_sample_classes`` would be ``>1``, it could still lead to only one *unique* label being sampled. >>> aug = iaa.BlendAlphaBoundingBoxes(None, >>> background=iaa.Multiply(0.0)) Create an augmenter that zeros all pixels (``Multiply(0.0)``) that are *not* (``background`` branch) within bounding boxes of *any* (``None``) label. In other words, all pixels outside of bounding boxes become black. Note that we don't use ``TotalDropout`` here, because by default it will also remove all coordinate-based augmentables, which will break the blending of such inputs. NrIc r)N)labelsnb_sample_labelsr)rVrrWBoundingBoxesMaskGen) r]rr<r>rrKr&rCrBr^rrrW rz BlendAlphaBoundingBoxes.__init__rrrrr^rrsqrc@seZdZdZdddZdS)IBatchwiseMaskGeneratora3Interface for classes generating masks for batches. Child classes are supposed to receive a batch and generate an iterable of masks, one per row (i.e. image), matching the row shape (i.e. image shape). This is used in :class:`~imgaug.augmenters.blend.BlendAlphaMask`. Added in 0.4.0. NcCdS)aGenerate a mask with given shape. Parameters ---------- batch : imgaug.augmentables.batches._BatchInAugmentation Shape of the mask to sample. random_state : None or int or imgaug.random.RNG or numpy.random.Generator or numpy.random.BitGenerator or numpy.random.SeedSequence or numpy.random.RandomState, optional A seed or random number generator to use during the sampling process. If ``None``, the global RNG will be used. See also :func:`~imgaug.augmenters.meta.Augmenter.__init__` for a similar parameter with more details. Returns ------- iterable of ndarray Masks, one per row in the batch. Each mask must be a ``float32`` array in interval ``[0.0, 1.0]``. It must either have the same shape as the row (i.e. the image) or shape ``(H, W)`` if all channels are supposed to have the same mask. Nrr]r@rCrrrr- z"IBatchwiseMaskGenerator.draw_masksrw)rrrrrrrrrr  s r cs2eZdZdZfddZd ddZddZZS) raMask generator that queries stochastic parameters for mask values. This class receives batches for which to generate masks, iterates over the batch rows (i.e. images) and generates one mask per row. For a row with shape ``(H, W, C)`` (= image shape), it generates either a ``(H, W)`` mask (if ``per_channel`` is false-like) or a ``(H, W, C)`` mask (if ``per_channel`` is true-like). The ``per_channel`` is sampled per batch for each row/image. Added in 0.4.0. Parameters ---------- parameter : imgaug.parameters.StochasticParameter Stochastic parameter to draw mask samples from. Expected to return values in interval ``[0.0, 1.0]`` (not all stochastic parameters do that) and must be able to handle sampling shapes ``(H, W)`` and ``(H, W, C)`` (all stochastic parameters should do that). per_channel : bool or float or imgaug.parameters.StochasticParameter, optional Whether to use the same mask for all channels (``False``) or to sample a new mask for each channel (``True``). If this value is a float ``p``, then for ``p`` percent of all rows (i.e. images) `per_channel` will be treated as ``True``, otherwise as ``False``. cs&tt|||_t|d|_dS)NrU)rVrrWrrXr[rU)r]rrUr^rrrWf s  z#StochasticParameterMaskGen.__init__NcsB|}tjjt|fd}fddt||DS)[ See :func:`~imgaug.augmenters.blend.IBatchwiseMaskGenerator.draw_masks`. rccsg|] \}}||qSr _draw_mask)rar$Z per_channel_irCr]rrrbw z9StochasticParameterMaskGen.draw_masks..)rgiarandomRNGrUrhr`r)r]r@rCrprUrrrrm s  z%StochasticParameterMaskGen.draw_maskscCst|dks |dkr|jj||d}n |jj|dd|d}|jdkr@d|dkr1dks@nJdt|t|f|S)Nr r!rcrrzYExpected 'parameter' samples to be in the interval [0.0, 1.0]. Got min %.4f and max %.4f.)r`rrhrr1r(r2r3)r]r$rCrUrrrrr| s  z%StochasticParameterMaskGen._draw_maskrw)rrrrrWrrrrrr^rrG s   rcseZdZdZddddgddffdd Zdd d Zd dZddZeddZ eddZ eddZ eddZ eddZ ZS)raGenerator that produces masks based on some similar colors in images. This class receives batches for which to generate masks, iterates over the batch rows (i.e. images) and generates one mask per row. The mask contains high alpha values for some colors, while other colors get low mask values. Which colors are chosen is random. How wide or narrow the selection is (e.g. very specific blue tone or all blue-ish colors) is determined by the hyperparameters. The color selection method performs roughly the following steps: 1. Split the full color range of the hue in ``HSV`` into ``nb_bins`` bins (i.e. ``256/nb_bins`` different possible hue tones). 2. Shift the bins by ``rotation_deg`` degrees. (This way, the ``0th`` bin does not always start at exactly ``0deg`` of hue.) 3. Sample ``alpha`` values for each bin. 4. Repeat the ``nb_bins`` bins until there are ``256`` bins. 5. Smoothen the alpha values of neighbouring bins using a gaussian kernel. The kernel's ``sigma`` is derived from ``smoothness``. 6. Associate all hue values in the image with the corresponding bin's alpha value. This results in the alpha mask. .. note:: This mask generator will produce an ``AssertionError`` for batches that contain no images. Added in 0.4.0. **Supported dtypes**: See :func:`~imgaug.augmenters.color.change_colorspaces_`. Parameters ---------- nb_bins : int or tuple of int or list of int or imgaug.parameters.StochasticParameter, optional Number of bins. For ``B`` bins, each bin denotes roughly ``360/B`` degrees of colors in the hue channel. Lower values lead to a coarser selection of colors. Expected value range is ``[2, 256]``. * If ``int``: Exactly that value will be used for all images. * If ``tuple`` ``(a, b)``: A random value will be uniformly sampled per image from the discrete interval ``[a..b]``. * If ``list``: A random value will be picked per image from that list. * If ``StochasticParameter``: That parameter will be queried once per batch for ``(N,)`` values -- one per image. smoothness : number or tuple of number or list of number or imgaug.parameters.StochasticParameter, optional Strength of the 1D gaussian kernel applied to the sampled binwise alpha values. Larger values will lead to more similar grayscaling of neighbouring colors. Expected value range is ``[0.0, 1.0]``. * If ``number``: Exactly that value will be used for all images. * If ``tuple`` ``(a, b)``: A random value will be uniformly sampled per image from the interval ``[a, b]``. * If ``list``: A random value will be picked per image from that list. * If ``StochasticParameter``: That parameter will be queried once per batch for ``(N,)`` values -- one per image. alpha : number or tuple of number or list of number or imgaug.parameters.StochasticParameter, optional Parameter to sample binwise alpha blending factors from. Expected value range is ``[0.0, 1.0]``. Note that the alpha values will be smoothed between neighbouring bins. Hence, it is usually a good idea to set this so that the probability distribution peaks are around ``0.0`` and ``1.0``, e.g. via a list ``[0.0, 1.0]`` or a ``Beta`` distribution. It is not recommended to set this to a deterministic value, otherwise all bins and hence all pixels in the generated mask will have the same value. * If ``number``: Exactly that value will be used for all bins. * If ``tuple`` ``(a, b)``: A random value will be uniformly sampled per bin from the interval ``[a, b]``. * If ``list``: A random value will be picked per bin from that list. * If ``StochasticParameter``: That parameter will be queried once per batch for ``(N*B,)`` values -- one per image and bin. rotation_deg : number or tuple of number or list of number or imgaug.parameters.StochasticParameter, optional Rotiational shift of each bin as a fraction of ``360`` degrees. E.g. ``0.0`` will not shift any bins, while a value of ``0.5`` will shift by around ``180`` degrees. This shift is mainly used so that the ``0th`` bin does not always start at ``0deg``. Expected value range is ``[-360, 360]``. This parameter can usually be kept at the default value. * If ``number``: Exactly that value will be used for all images. * If ``tuple`` ``(a, b)``: A random value will be uniformly sampled per image from the interval ``[a, b]``. * If ``list``: A random value will be picked per image from that list. * If ``StochasticParameter``: That parameter will be queried once per batch for ``(N,)`` values -- one per image. from_colorspace : str, optional The source colorspace (of the input images). See :func:`~imgaug.augmenters.color.change_colorspace_`. rrrHrrrcsvtt|tj|ddddd|_tj|ddddd|_tj|ddddd|_tj|dd ddd|_ ||_ d |_ dS) Nr)rTrNrrGr6r)irg$@) rVrrWrXhandle_discrete_paramrrYrr6rr sigma_max)r]rrr6rrr^rrrW s& zSomeColorsMaskGen.__init__NcsD|jdus Jdt|}j||dfddt|jDS)r NzXCan only generate masks for batches that contain images, but got a batch without images.rccg|] \}}||qSrr)rarimagesamplesr]rrrb rz0SomeColorsMaskGen.draw_masks..)rerr _draw_samplesrir rrrr s   zSomeColorsMaskGen.draw_maskscCs2|||d||d||j|d||jS)Nrrr ) generate_maskrr)r]r image_idxrrrrr% s  zSomeColorsMaskGen._draw_maskc Cs|j}|jj|f|d}|jj|f|d}|jjt|f|d}|jj|f|d}t|dd}t|dd}t|dd}t t |d tj d}t ||} | ||fS)NrcrrrHrgll?)rrrhrr6r(sumrr"modr#r*Zint32r) r]r@rCrrrr6r rotation_binsbinwise_alphasrrrr. s,    zSomeColorsMaskGen._draw_samplesc Csddlm}|jt||j|d}d|jddvr(tj|jddtjdS| |}| ||}| ||}| ||} | S)aGGenerate a colorwise alpha mask for a single image. Added in 0.4.0. Parameters ---------- image : ndarray Image for which to generate the mask. Must have shape ``(H,W,3)`` in colorspace `from_colorspace`. binwise_alphas : ndarray Alpha values of shape ``(B,)`` with ``B`` in ``[1, 256]`` and values in interval ``[0.0, 1.0]``. Will be upscaled to 256 bins by simple repetition. Each bin represents ``1/256`` th of the hue. sigma : float Sigma of the 1D gaussian kernel applied to the upscaled binwise alpha value array. rotation_bins : int By how much to rotate the 256 bin alpha array. The rotation is given in number of bins. from_colorspace : str Colorspace of the input image. One of ``imgaug.augmenters.color.CSPACE_*``. Returns ------- ndarray ``float32`` mask array of shape ``(H, W)`` with values in ``[0.0, 1.0]`` r)color)Z to_colorspacerrr r) r"Zchange_colorspace_r(r0Z CSPACE_HSVr$rr+_upscale_to_256_alpha_bins_rotate_alpha_bins_smoothen_alphas_generate_pixelwise_alpha_mask) rrr!sigmar rZcolorlib image_hsvZbinwise_alphas_smoothrrrrrD s '   zSomeColorsMaskGen.generate_maskcCs8t|}ttd|}t||f}|dd}|S)Nrr)r`rr(ceilrepeat)rrrrZnb_repeats_per_binrrrr$ s  z,SomeColorsMaskGen._upscale_to_256_alpha_binscCs|dkr t|| }|Sr)r(Zroll)rrrr rrrr% sz$SomeColorsMaskGen._rotate_alpha_binscCs|dkr|Stt|dd}d|}}|ddkr|d7}t|| d||d|g}tjt|tjddf||f||tjddddf}||| }|S)Nrg@rrr r)ksizeZsigmaXZsigmaYZ borderType) r3rr( concatenatecv2Z GaussianBlurrr)ZBORDER_REPLICATE)rrrr(r,Zksize_yZksize_xrrrr& s*     z"SomeColorsMaskGen._smoothen_alphascCsV|dddddf}|d}tt|ddtj}t||}|tjdS)Nrgo@)r(r"r#r*Zuint8rZ apply_lutr+)rr)Z hue_to_alphahuetablerrrrr' s  z0SomeColorsMaskGen._generate_pixelwise_alpha_maskrw)rrrrrWrrrrrr$r%r&r'rrrr^rr s&g   :   rc@sPeZdZ  dddZdddZdd Zd d Zeed d Z eddZ dS)_LinearGradientMaskGenrHrcCsb||_tj|ddddd|_tj|ddddd|_tj|ddddd|_tj|ddddd|_dS)NrrGTrNrrr)rrXrYrrrr)r]rrrrrrrrrW s" z_LinearGradientMaskGen.__init__Ncs<t|}|}jt||dfddt|DS)t See :func:`~imgaug.augmenters.blend.IBatchwiseMaskGenerator.draw_masks`. Added in 0.4.0. rccrrr)rarr$rrrrb rz5_LinearGradientMaskGen.draw_masks..)rrrgrr`ri)r]r@rCrprrrr s  z!_LinearGradientMaskGen.draw_maskscCs2|||d||d||d||d|S)Nrrr rr)r]r$rrrrrr s    z!_LinearGradientMaskGen._draw_maskcCsT|jj|f|d}|jj|f|d}|jj|f|d}|jj|f|d}||||fSNrc)rrhrrr)r]rrCrrrrrrrr s   z$_LinearGradientMaskGen._draw_samplescCr )aGenerate a horizontal gradient mask. Added in 0.4.0. Parameters ---------- shape : tuple of int Shape of the image. The mask will have the same height and width. min_value : number Minimum value of the gradient in interval ``[0.0, 1.0]``. max_value : number Maximum value of the gradient in interval ``[0.0, 1.0]``. start_at : number Position on the x-axis where the linear gradient starts, given as a fraction of the axis size. Interval is ``[0.0, 1.0]``. end_at : number Position on the x-axis where the linear gradient ends, given as a fraction of the axis size. Interval is ``[0.0, 1.0]``. Returns ------- ndarray ``float32`` mask array with same height and width as the image. Values are in ``[0.0, 1.0]``. Nrrr$rrrrrrrr r z$_LinearGradientMaskGen.generate_maskcCs:|dd\}}||} tt|dd}tt|dd}ttt|| d| } ttt|| d| } d} | | krAd} | | } } tj| f|tjd} tj||| | tjd}tj| | f|tjd}tj| ||fdd }| rsd|}|dkr|ddtjf}t |d |f}|S|tjddf}t ||d f}|S) Nrr rHrFTr)startstopnumrrr) r2r3rr(fullr+Zlinspacer-r)r.)rr$rrrrrheightwidthZ axis_sizeZ start_at_pxZ end_at_pxinvertedZ before_gradZgradZ after_gradrrrr_generate_mask sJ  z%_LinearGradientMaskGen._generate_mask)rHrrHrrw) rrrrWrrrrrrr>rrrrr2 s    "r2c2eZdZdZ  dfdd ZeddZZS) ra2 Generator that produces horizontal linear gradient masks. This class receives batches and produces for each row (i.e. image) a horizontal linear gradient that matches the row's shape (i.e. image shape). The gradient increases linearly from a minimum value to a maximum value along the x-axis. The start and end points (i.e. where the minimum value starts to increase and where it reaches the maximum) may be defines as fractions of the width. E.g. for width ``100`` and ``start=0.25``, ``end=0.75``, the gradient would have its minimum in interval ``[0px, 25px]`` and its maximum in interval ``[75px, 100px]``. Note that this has nothing to do with a *derivative* along the x-axis. Added in 0.4.0. Parameters ---------- min_value : number or tuple of number or list of number or imgaug.parameters.StochasticParameter, optional Minimum value that the mask will have up to the start point of the linear gradient. Note that `min_value` is allowed to be larger than `max_value`, in which case the gradient will start at the (higher) `min_value` and decrease towards the (lower) `max_value`. * If ``number``: Exactly that value will be used for all images. * If ``tuple`` ``(a, b)``: A random value will be uniformly sampled per image from the interval ``[a, b]``. * If ``list``: A random value will be picked per image from that list. * If ``StochasticParameter``: That parameter will be queried once per batch for ``(N,)`` values -- one per image. max_value : number or tuple of number or list of number or imgaug.parameters.StochasticParameter, optional Maximum value that the mask will have at the end of the linear gradient. Datatypes are analogous to `min_value`. start_at : number or tuple of number or list of number or imgaug.parameters.StochasticParameter, optional Position on the x-axis where the linear gradient starts, given as a fraction of the axis size. Interval is ``[0.0, 1.0]``, where ``0.0`` is at the left of the image. If ``end_at < start_at`` the gradient will be inverted. Datatypes are analogous to `min_value`. end_at : number or tuple of number or list of number or imgaug.parameters.StochasticParameter, optional Position on the x-axis where the linear gradient ends, given as a fraction of the axis size. Interval is ``[0.0, 1.0]``, where ``0.0`` is at the right of the image. Datatypes are analogous to `min_value`. rrctt|jd||||ddS)Nrrrrrr)rVrrWr]rrrrr^rrrW|   z(HorizontalLinearGradientMaskGen.__init__cC|jd|||||dS)Generate a linear horizontal gradient mask. Added in 0.4.0. Parameters ---------- shape : tuple of int Shape of the image. The mask will have the same height and width. min_value : number Minimum value of the gradient in interval ``[0.0, 1.0]``. max_value : number Maximum value of the gradient in interval ``[0.0, 1.0]``. start_at : number Position on the x-axis where the linear gradient starts, given as a fraction of the axis size. Interval is ``[0.0, 1.0]``. end_at : number Position on the x-axis where the linear gradient ends, given as a fraction of the axis size. Interval is ``[0.0, 1.0]``. Returns ------- ndarray ``float32`` mask array with same height and width as the image. Values are in ``[0.0, 1.0]``. rrr$rrrrr>r6rrrr !z-HorizontalLinearGradientMaskGen.generate_maskrrrrrrrrrWrrrrrr^rrD s7 rcr?) raGenerator that produces vertical linear gradient masks. See :class:`~imgaug.augmenters.blend.HorizontalLinearGradientMaskGen` for details. Added in 0.4.0. Parameters ---------- min_value : number or tuple of number or list of number or imgaug.parameters.StochasticParameter, optional Minimum value that the mask will have up to the start point of the linear gradient. Note that `min_value` is allowed to be larger than `max_value`, in which case the gradient will start at the (higher) `min_value` and decrease towards the (lower) `max_value`. * If ``number``: Exactly that value will be used for all images. * If ``tuple`` ``(a, b)``: A random value will be uniformly sampled per image from the interval ``[a, b]``. * If ``list``: A random value will be picked per image from that list. * If ``StochasticParameter``: That parameter will be queried once per batch for ``(N,)`` values -- one per image. max_value : number or tuple of number or list of number or imgaug.parameters.StochasticParameter, optional Maximum value that the mask will have at the end of the linear gradient. Datatypes are analogous to `min_value`. start_at : number or tuple of number or list of number or imgaug.parameters.StochasticParameter, optional Position on the y-axis where the linear gradient starts, given as a fraction of the axis size. Interval is ``[0.0, 1.0]``, where ``0.0`` is at the top of the image. If ``end_at < start_at`` the gradient will be inverted. Datatypes are analogous to `min_value`. end_at : number or tuple of number or list of number or imgaug.parameters.StochasticParameter, optional Position on the x-axis where the linear gradient ends, given as a fraction of the axis size. Interval is ``[0.0, 1.0]``, where ``1.0`` is at the bottom of the image. Datatypes are analogous to `min_value`. rrcr@)NrrA)rVrrWrBr^rrrW rCz&VerticalLinearGradientMaskGen.__init__cCrD)rErrFrGr6rrrr rHz+VerticalLinearGradientMaskGen.generate_maskrIrJrrr^rr s/ rc@s>eZdZdZddgfddZd ddZd d Zed d ZdS)raGenerate masks following a regular grid pattern. This mask generator splits each image into a grid-like pattern of ``H`` rows and ``W`` columns. Each cell is then filled with an alpha value, sampled randomly per cell. The difference to :class:`CheckerboardMaskGen` is that this mask generator samples random alpha values per cell, while in the checkerboard the alpha values follow a fixed pattern. Added in 0.4.0. Parameters ---------- nb_rows : int or tuple of int or list of int or imgaug.parameters.StochasticParameter Number of rows of the regular grid. * If ``int``: Exactly that value will be used for all images. * If ``tuple`` ``(a, b)``: A random value will be uniformly sampled per image from the discrete interval ``[a..b]``. * If ``list``: A random value will be picked per image from that list. * If ``StochasticParameter``: That parameter will be queried once per batch for ``(N,)`` values -- one per image. nb_cols : int or tuple of int or list of int or imgaug.parameters.StochasticParameter Number of columns of the checkerboard. Analogous to `nb_rows`. alpha : number or tuple of number or list of number or imgaug.parameters.StochasticParameter, optional Alpha value of each cell. * If ``number``: Exactly that value will be used for all images. * If ``tuple`` ``(a, b)``: A random value will be uniformly sampled per image from the interval ``[a, b]``. * If ``list``: A random value will be picked per image from that list. * If ``StochasticParameter``: That parameter will be queried once per batch for ``(N,)`` values -- one per image. rHrcCsJtj|dddddd|_tj|dddddd|_tj|ddddd |_dS) Nr)rNTFrOrPrQZ allow_floatsrr6rGrN)rXrrrrYr6)r]rrr6rrrrW< s zRegularGridMaskGen.__init__NcsFt|}|}jt||d\}}}fddt||||DS)r3rccs$g|]\}}}}||||qSrr4)rar$ nb_rows_i nb_cols_iZalpha_iryrrrbV s z1RegularGridMaskGen.draw_masks..)rrrgrr`r)r]r@rCrprrr6rryrrJ s    zRegularGridMaskGen.draw_maskscCsX|jj|f|d}|jj|f|d}||}|jjt|f|d}t||}|||fSr5)rrhrr6r(rr)r]rqrCrrZnb_alphas_per_imgZ alpha_rawr6rrrr[ s     z RegularGridMaskGen._draw_samplescCshddlm}|dd\}}d||fvrtj||ftjdStt|d|}tt|d|}t||}t||} |jd||}|j||ksZJd|j||||||ff| tj ||f} tj | |dd} tj | | dd} || j d} || j d} tt | d} tt| d}tt | d}tt| d}|j| | |||dd } | S) aGenerate a mask following a checkerboard pattern. Added in 0.4.0. Parameters ---------- shape : tuple of int Height and width of the output mask. nb_rows : int Number of rows of the checkerboard pattern. nb_cols : int Number of columns of the checkerboard pattern. alphas : ndarray 1D or 2D array containing for each cell the alpha value, i.e. ``nb_rows*nb_cols`` values. Returns ------- ndarray ``float32`` mask array with same height and width as ``segmap.shape``. Values are in ``[0.0, 1.0]``. r)rrr rzExpected `alphas` to not contain less values than `nb_rows * nb_cols` (both clipped to [1, height] and [1, width] respectively). Got %d alpha values vs %d expected values (nb_rows=%d, nb_cols=%d) for requested mask shape %s.rZreflect)toprightbottomleftr)r#rr(rr+r2r3rZflatr*r-r+r$floorr*pad)rr$rrrrZsizelibr;r<Z cell_height cell_widthrZmissing_heightZ missing_widthrNrPrQrOrrrri s<    z RegularGridMaskGen.generate_maskrw rrrrrWrrrrrrrrr s) rc@sFeZdZdZddZeddZeddZd d d Ze d d Z dS)raGenerate masks following a checkerboard-like pattern. This mask generator splits each image into a regular grid of ``H`` rows and ``W`` columns. Each cell is then filled with either ``1.0`` or ``0.0``. The cell at the top-left is always ``1.0``. Its right and bottom neighbour cells are ``0.0``. The 4-neighbours of any cell always have a value opposite to the cell's value (``0.0`` vs. ``1.0``). Added in 0.4.0. Parameters ---------- nb_rows : int or tuple of int or list of int or imgaug.parameters.StochasticParameter, optional Number of rows of the checkerboard. * If ``int``: Exactly that value will be used for all images. * If ``tuple`` ``(a, b)``: A random value will be uniformly sampled per image from the discrete interval ``[a..b]``. * If ``list``: A random value will be picked per image from that list. * If ``StochasticParameter``: That parameter will be queried once per batch for ``(N,)`` values -- one per image. nb_cols : int or tuple of int or list of int or imgaug.parameters.StochasticParameter, optional Number of columns of the checkerboard. Analogous to `nb_rows`. cCst||dd|_dS)Nrr)rgrid)r]rrrrrrW s zCheckerboardMaskGen.__init__cCr)zGet the number of rows of the checkerboard grid. Added in 0.4.0. Returns ------- int The number of rows. )rVrryrrrr  zCheckerboardMaskGen.nb_rowscCr)zGet the number of columns of the checkerboard grid. Added in 0.4.0. Returns ------- int The number of columns. )rVrryrrrr rWzCheckerboardMaskGen.nb_colsNcsFt|}|}jjt||d\}}}fddt|||DS)r3rccs g|] \}}}|||qSrr4)rar$rLrMryrrrb sz2CheckerboardMaskGen.draw_masks..)rrrgrVrr`r)r]r@rCrprrZ_alpharryrr s   zCheckerboardMaskGen.draw_maskscCs|dd\}}d||fvrtj||ftjdStt|d|}tt|d|}tj|fdtjd}d|ddd<t|tjddf|df}d|dddddf|dddddf<t ||||S)aGenerate a mask following a checkerboard pattern. Added in 0.4.0. Parameters ---------- shape : tuple of int Height and width of the output mask. nb_rows : int Number of rows of the checkerboard pattern. nb_cols : int Number of columns of the checkerboard pattern. Returns ------- ndarray ``float32`` mask array with same height and width as ``segmap.shape``. Values are in ``[0.0, 1.0]``. rr rrrrHN) r(rr+r2r3r:r.r)rr)rr$rrr;r<rrrrrr s ,z!CheckerboardMaskGen.generate_maskrw) rrrrrWrrrrrrrrrrr s   rc@s8eZdZdZd ddZd ddZddZed d ZdS) ras Generator that produces masks highlighting segmentation map classes. This class produces for each segmentation map in a batch a mask in which the locations of a set of provided classes are highlighted (i.e. ``1.0``). The classes may be provided as a fixed list of class ids or a stochastic parameter from which class ids will be sampled. The produced masks are initially of the same height and width as the segmentation map arrays and later upscaled to the image height and width. .. note:: Segmentation maps can have multiple channels. If that is the case then for each position ``(x, y)`` it is sufficient that any class id in any channel matches one of the desired class ids. .. note:: This class will produce an ``AssertionError`` if there are no segmentation maps in a batch. Added in 0.4.0. Parameters ---------- class_ids : int or tuple of int or list of int or imgaug.parameters.StochasticParameter Segmentation map classes to mark in the produced mask. If `nb_sample_classes` is ``None`` then this is expected to be either a single ``int`` (always mark this one class id) or a ``list`` of ``int`` s (always mark these class ids). If `nb_sample_classes` is set, then this parameter will be treated as a stochastic parameter with the following valid types: * If ``int``: Exactly that class id will be used for all segmentation maps. * If ``tuple`` ``(a, b)``: ``N`` random values will be uniformly sampled per segmentation map from the discrete interval ``[a..b]`` and used as the class ids. * If ``list``: ``N`` random values will be picked per segmentation map from that list and used as the class ids. * If ``StochasticParameter``: That parameter will be queried once per batch for ``(sum(N),)`` values. ``N`` denotes the number of classes to sample per segmentation map (derived from `nb_sample_classes`) and ``sum(N)`` denotes the sum of ``N`` s over all segmentation maps. nb_sample_classes : None or tuple of int or list of int or imgaug.parameters.StochasticParameter, optional Number of class ids to sample (with replacement) per segmentation map. As sampling happens with replacement, fewer *unique* class ids may be sampled. * If ``None``: `class_ids` is expected to be a fixed value of class ids to be used for all segmentation maps. * If ``int``: Exactly that many class ids will be sampled for all segmentation maps. * If ``tuple`` ``(a, b)``: A random value will be uniformly sampled per segmentation map from the discrete interval ``[a..b]``. * If ``list`` or ``int``: A random value will be picked per segmentation map from that list. * If ``StochasticParameter``: That parameter will be queried once per batch for ``(B,)`` values, where ``B`` is the number of segmentation maps. NcCsz|dur#t|r |g}t|tsJdt|jf||_d|_dStj |dddddd|_tj |dddddd|_dS)NzExpected `class_ids` to be a single integer or a list of integers if `nb_sample_classes` is None. Got type `%s`. Set `nb_sample_classes` to e.g. an integer to enable stochastic parameters for `class_ids`.rrNTFrKr) rZis_single_integer isinstancelisttyperrrrXr)r]rrrrrrWi s*    zSegMapClassIdsMaskGen.__init__csF|jdus Jdt|}j|j|d}fddt|j|DS)r3NzaCan only generate masks for batches that contain segmentation maps, but got a batch without them.rccg|] \}}||qSrr4)rasegmapZ class_ids_iryrrrb z4SegMapClassIdsMaskGen.draw_masks..)rrrrrr)r]r@rCrrryrr s    z SegMapClassIdsMaskGen.draw_maskscC|j}|durt|jtsJdt|jjf|jg|S|j|f|d}t|dd}|jjt |f|d}t ||}|SNzExpected list got %s.rcr) rrYrrZr[rrhr(r"rr)r]rrCrZ class_ids_rawrrrrr    z#SegMapClassIdsMaskGen._draw_samplescCsjtj|jjddtd}|D]}tj|j|kdd}t||}q|tj}t ||jdd}|S)aGenerate a mask of where the segmentation map has the given classes. Added in 0.4.0. Parameters ---------- segmap : imgaug.augmentables.segmap.SegmentationMapsOnImage The segmentation map for which to generate the mask. class_ids : iterable of int IDs of the classes to set to ``1.0``. For an ``(x, y)`` position, it is enough that *any* channel at the given location to have one of these class ids to be marked as ``1.0``. Returns ------- ndarray ``float32`` mask array with same height and width as ``segmap.shape``. Values are in ``[0.0, 1.0]``. rr rr) r(rrr$boolany logical_orr*r+rr)rr]rrZclass_idZmask_irrrr s z#SegMapClassIdsMaskGen.generate_maskrwrUrrrrr" s F rc@s8eZdZdZd ddZd ddZddZed d ZdS) ra] Generator that produces masks highlighting bounding boxes. This class produces for each row (i.e. image + bounding boxes) in a batch a mask in which the inner areas of bounding box rectangles with given labels are marked (i.e. set to ``1.0``). The labels may be provided as a fixed list of strings or a stochastic parameter from which labels will be sampled. If no labels are provided, all bounding boxes will be marked. A pixel will be set to ``1.0`` if *at least* one bounding box at that location has one of the requested labels, even if there is *also* one bounding box at that location with a not requested label. .. note:: This class will produce an ``AssertionError`` if there are no bounding boxes in a batch. Added in 0.4.0. Parameters ---------- labels : None or str or list of str or imgaug.parameters.StochasticParameter Labels of bounding boxes to select for. If `nb_sample_labels` is ``None`` then this is expected to be either also ``None`` (select all BBs) or a single ``str`` (select BBs with this one label) or a ``list`` of ``str`` s (always select BBs with these labels). If `nb_sample_labels` is set, then this parameter will be treated as a stochastic parameter with the following valid types: * If ``None``: Ignore the sampling count and always use all bounding boxes. * If ``str``: Exactly that label will be used for all images. * If ``list`` of ``str``: ``N`` random values will be picked per image from that list and used as the labels. * If ``StochasticParameter``: That parameter will be queried once per batch for ``(sum(N),)`` values. ``N`` denotes the number of labels to sample per segmentation map (derived from `nb_sample_labels`) and ``sum(N)`` denotes the sum of ``N`` s over all images. nb_sample_labels : None or tuple of int or list of int or imgaug.parameters.StochasticParameter, optional Number of labels to sample (with replacement) per image. As sampling happens with replacement, fewer *unique* labels may be sampled. * If ``None``: `labels` is expected to also be ``None`` or a fixed value of labels to be used for all images. * If ``int``: Exactly that many labels will be sampled for all images. * If ``tuple`` ``(a, b)``: A random value will be uniformly sampled per image from the discrete interval ``[a..b]``. * If ``list``: A random value will be picked per image from that list. * If ``StochasticParameter``: That parameter will be queried once per batch for ``(B,)`` values, where ``B`` is the number of images. NcCs|dur d|_d|_dS|dur/t|r|g}t|ts'Jdt|jf||_d|_dSt |d|_tj |dddddd|_dS)NzExpected `labels` a single string or a list of strings if `nb_sample_labels` is None. Got type `%s`. Set `nb_sample_labels` to e.g. an integer to enable stochastic parameters for `labels`.rrrXTFrK) rrrZ is_stringrYrZr[rrXZhandle_categorical_string_paramr)r]rrrrrrW s(     zBoundingBoxesMaskGen.__init__csd|jdus Jdt|}jdurfdd|jDSj|j|d}fddt|j|DS)r3Nz^Can only generate masks for batches that contain bounding boxes, but got a batch without them.csg|]}|dqSrwr4)rabbsoiryrrrb0sz3BoundingBoxesMaskGen.draw_masks..rccr\rr4)rareZlabels_iryrrrb5r^)rrrrrrr)r]r@rCrrryrr#s      zBoundingBoxesMaskGen.draw_maskscCr_r`) rrYrrZr[rrhr(r"rr)r]rrCrZ labels_rawrrrrr:raz"BoundingBoxesMaskGen._draw_samplesc Cs|durt|nd}|jdd\}}tj||ftjd}|D]I}|dus*|j|vrhttt|j d|}ttt|j d|}ttt|j d|} ttt|j d|} || krh|| krhd||| || f<q|S)aGenerate a mask of the areas of bounding boxes with given labels. Added in 0.4.0. Parameters ---------- bbsoi : imgaug.augmentables.bbs.BoundingBoxesOnImage The bounding boxes for which to generate the mask. labels : None or iterable of str Labels of the bounding boxes to set to ``1.0``. For an ``(x, y)`` position, it is enough that *any* bounding box at the given location has one of the labels. If this is ``None``, all bounding boxes will be marked. Returns ------- ndarray ``float32`` mask array with same height and width as ``segmap.shape``. Values are in ``[0.0, 1.0]``. Nrr rr) setr$r(rr+labelr2r3rx1y1x2y2) rrerr;r<rZbbrhrirjrkrrrrNsz"BoundingBoxesMaskGen.generate_mask)NNrwrUrrrrr s A rc@s"eZdZdZddZdddZdS) InvertMaskGena5Generator that inverts the outputs of other mask generators. This class receives batches and calls for each row (i.e. image) a child mask generator to produce a mask. That mask is then inverted for ``p%`` of all rows, i.e. converted to ``1.0 - mask``. Added in 0.4.0. Parameters ---------- p : bool or float or imgaug.parameters.StochasticParameter, optional Probability of inverting each mask produced by the other mask generator. child : IBatchwiseMaskGenerator The other mask generator to invert. cCst|d|_||_dS)Nr)rXr[rchild)r]rrmrrrrWs zInvertMaskGen.__init__NcCsZt|}|jj||d}|jjt||d}t||D]\}}|dkr*d||d<q|S)r3rcr!r.)rrrmrrrhr`r)r]r@rCrrrZp_irrrrs  zInvertMaskGen.draw_masksrw)rrrrrWrrrrrrlvsrlAlphazAlpha is deprecated. Use BlendAlpha instead. The order of parameters is the same. Parameter 'first' was renamed to 'foreground'. Parameter 'second' was renamed to 'background'.)Zalt_funccommentFrIc Ct||||||||dS)z;See :class:`BlendAlpha`. Deprecated since 0.4.0. r)rFrLfirstsecondrUrKr&rCrBrrrrnAlphaElementwisezAlphaElementwise is deprecated. Use BlendAlphaElementwise instead. The order of parameters is the same. Parameter 'first' was renamed to 'foreground'. Parameter 'second' was renamed to 'background'.c Crp)zFSee :class:`BlendAlphaElementwise`. Deprecated since 0.4.0. r)rrqrrrrurtzSimplexNoiseAlpha is deprecated. Use BlendAlphaSimplexNoise instead. The order of parameters is the same. Parameter 'first' was renamed to 'foreground'. Parameter 'second' was renamed to 'background'.rrr3Tc Cs"t|||||||||| | | | d S)zGSee :class:`BlendAlphaSimplexNoise`. Deprecated since 0.4.0. ) r<r>rUrrrrrrrKr&rCrB)r) rrrsrUrrrrrrrKr&rCrBrrrSimplexNoiseAlphasrvzFrequencyNoiseAlpha is deprecated. Use BlendAlphaFrequencyNoise instead. The order of parameters is the same. Parameter 'first' was renamed to 'foreground'. Parameter 'second' was renamed to 'background'.rrrr!cCs$t|||||||||| | | | | dS)zISee :class:`BlendAlphaFrequencyNoise`. Deprecated since 0.4.0. )rr<r>rUrrrrrrrKr&rCrB)r)rrrrsrUrrrrrrrKr&rCrBrrrFrequencyNoiseAlphasrw)r)rNNFNNrIrIr)>r __future__rrrabcrrnumpyr(sixr.ZimgaugrZ imgaug.imgaugrr#r r rXr r4r rZ augmentablesrrrr9rArEZ AugmenterrFrrrrrrrrrrrZ add_metaclassobjectr rrr2rrrrrrrlrIrnrurvrwrrrrs       *t;a}ze&O( kcu*,)