U h4@sddlmZddlZddlmZmZmZmZmZm Z ddl Z ddl Z ddl mZmZmZddlmZddlmZmZddlmZmZmZddlmZd d lmZd d d dgZGdd d eZ GdddeZ!Gdd d eZ"Gdd d eZ#GdddeZ$dS)) annotationsN)AnyListSequenceTupleUnioncast)FieldValidationInfofield_validator)InterpolationType)BaseTransformInitSchema DualTransform)ScaleFloatType ScaleIntTypeTargetsto_tuple) functional RandomScaleLongestMaxSizeSmallestMaxSizeResizecseZdZdZejejejejfZ Gddde Z de j ddfddd d d fd d ZddddZdd dddddZdd ddddddZddddddZdd dddddZd dd!d"ZZS)#rauRandomly resize the input. Output image size is different from the input image size. Args: scale_limit (float or tuple[float, float]): scaling factor range. If scale_limit is a single float value, the range will be (-scale_limit, scale_limit). Note that the scale_limit will be biased by 1. If scale_limit is a tuple, like (low, high), sampling will be done from the range (1 + low, 1 + high). Default: (-0.1, 0.1). interpolation (OpenCV flag): flag that is used to specify the interpolation algorithm. Should be one of: cv2.INTER_NEAREST, cv2.INTER_LINEAR, cv2.INTER_CUBIC, cv2.INTER_AREA, cv2.INTER_LANCZOS4. Default: cv2.INTER_LINEAR. p (float): probability of applying the transform. Default: 0.5. Targets: image, mask, bboxes, keypoints Image types: uint8, float32 Note: - The output image size is different from the input image size. - Scale factor is sampled independently per image side (width and height). - Bounding box coordinates are scaled accordingly. - Keypoint coordinates are scaled accordingly. Mathematical formulation: Let (W, H) be the original image dimensions and (W', H') be the output dimensions. The scale factor s is sampled from the range [1 + scale_limit[0], 1 + scale_limit[1]]. Then, W' = W * s and H' = H * s. Example: >>> import numpy as np >>> import albumentations as A >>> image = np.random.randint(0, 256, (100, 100, 3), dtype=np.uint8) >>> transform = A.RandomScale(scale_limit=0.1, p=1.0) >>> result = transform(image=image) >>> scaled_image = result['image'] # scaled_image will have dimensions in the range [90, 110] x [90, 110] # (assuming the scale_limit of 0.1 results in a scaling factor between 0.9 and 1.1) c@s:eZdZUded<ded<ededddddZd S) zRandomScale.InitSchemar scale_limitr interpolationztuple[float, float])vreturncCs t|ddS)Ng?Zbiasr)clsrr Q/tmp/pip-unpacked-wheel-e8onvpoz/albumentations/augmentations/geometric/resize.pycheck_scale_limitDsz(RandomScale.InitSchema.check_scale_limitN__name__ __module__ __qualname____annotations__r classmethodr"r r r r! InitSchema@s r))gg?Ng?rint bool | Nonefloat)rr always_applypcs.tj||dttttf||_||_dSN)r.r-)super__init__rrr,rr)selfrrr-r. __class__r r!r1IszRandomScale.__init__zdict[str, float]rcCsdtj|jiS)Nscale)randomuniformrr2r r r! get_paramsTszRandomScale.get_params np.ndarrayr)imgr6paramsrcKst|||jSN) fgeometricr6r)r2r<r6r=r r r!applyWszRandomScale.apply)maskr6rr=rcKst||tjSr>)r?r6cv2 INTER_NEAREST)r2rAr6rr=r r r! apply_to_mask_szRandomScale.apply_to_maskbboxesr=rcKs|Sr>r r2rFr=r r r!apply_to_bboxeshszRandomScale.apply_to_bboxes) keypointsr6r=rcKst|||Sr>)r?keypoints_scale)r2rIr6r=r r r!apply_to_keypointslszRandomScale.apply_to_keypointszdict[str, Any]cCs|jt|jdddS)Ngr)rr)rrrr9r r r!get_transform_init_argstsz#RandomScale.get_transform_init_args)r$r%r&__doc__rIMAGEMASKBBOXES KEYPOINTS_targetsr r)rB INTER_LINEARr1r:r@rDrHrKrL __classcell__r r r3r!rs)   c@s<eZdZUded<ded<ededddddd Zd S) MaxSizeInitSchemazint | list[int]max_sizer rrr )rinforcCsPt|ttfr|n|g}|D]}|dkst|jdqtttttf|S)Nrz must be bigger or equal to 1.) isinstancelisttuple ValueError field_namerrr*r)rrrWresultvaluer r r!r"|s z#MaxSizeInitSchema.check_scale_limitNr#r r r r!rUxs rUcseZdZdZejejejejfZ Gddde Z de j ddfddd d d fd d ZddddddddZddddddZdddddddZddddZddddZZS) raRescale an image so that the longest side is equal to max_size, keeping the aspect ratio of the initial image. Args: max_size (int, Sequence[int]): Maximum size of the image after the transformation. When using a list or tuple, the max size will be randomly selected from the values provided. interpolation (OpenCV flag): interpolation method. Default: cv2.INTER_LINEAR. p (float): probability of applying the transform. Default: 1. Targets: image, mask, bboxes, keypoints Image types: uint8, float32 Note: - If the longest side of the image is already less than or equal to max_size, the image will not be resized. - This transform will not crop the image. The resulting image may be smaller than max_size in both dimensions. - For non-square images, the shorter side will be scaled proportionally to maintain the aspect ratio. Example: >>> import albumentations as A >>> import cv2 >>> transform = A.Compose([ ... A.LongestMaxSize(max_size=1024, interpolation=cv2.INTER_LINEAR), ... A.PadIfNeeded(min_height=1024, min_width=1024, border_mode=cv2.BORDER_CONSTANT), ... ]) >>> # Assume we have a 1500x800 image >>> transformed = transform(image=image) >>> transformed_image = transformed['image'] >>> transformed_image.shape (1024, 546, 3) # The longest side (1500) is scaled to 1024, and the other side is scaled proportionally c@s eZdZdS)zLongestMaxSize.InitSchemaNr$r%r&r r r r!r)sr)Nrint | Sequence[int]r*r+r,rVrr-r.cst||||_||_dSr>r0r1rrVr2rVrr-r.r3r r!r1szLongestMaxSize.__init__r;rr<rVrr=rcKstj|||dSN)rVr)r?Zlongest_max_sizer2r<rVrr=r r r!r@szLongestMaxSize.applyrEcKs|Sr>r rGr r r!rHszLongestMaxSize.apply_to_bboxesrIrVr=rcKs*|ddd}|t|}t|||SNshape)maxr?rJr2rIrVr=Z image_shaper6r r r!rKs z!LongestMaxSize.apply_to_keypointsdict[str, int]r5cCs"dt|jtr|jn t|jiSNrVrXrVr*r7choicer9r r r!r:szLongestMaxSize.get_paramstuple[str, ...]cCsdSrfr r9r r r!get_transform_init_args_namessz,LongestMaxSize.get_transform_init_args_names)r$r%r&rMrrNrOrPrQrRrUr)rBrSr1r@rHrKr:rsrTr r r3r!rs"   cseZdZdZejejejejfZ Gddde Z de j ddfddd d d fd d ZddddddddZddddddZdddddddZddddZddddZZS) raRescale an image so that minimum side is equal to max_size, keeping the aspect ratio of the initial image. Args: max_size (int, list of int): Maximum size of smallest side of the image after the transformation. When using a list, max size will be randomly selected from the values in the list. interpolation (OpenCV flag): Flag that is used to specify the interpolation algorithm. Should be one of: cv2.INTER_NEAREST, cv2.INTER_LINEAR, cv2.INTER_CUBIC, cv2.INTER_AREA, cv2.INTER_LANCZOS4. Default: cv2.INTER_LINEAR. p (float): Probability of applying the transform. Default: 1. Targets: image, mask, bboxes, keypoints Image types: uint8, float32 Note: - If the smallest side of the image is already less than or equal to max_size, the image will not be resized. - This transform will not crop the image. The resulting image may be larger than max_size in both dimensions. - For non-square images, the larger side will be scaled proportionally to maintain the aspect ratio. - Bounding boxes and keypoints are scaled accordingly. Mathematical Details: 1. Let (W, H) be the original width and height of the image. 2. The scaling factor s is calculated as: s = max_size / min(W, H) 3. The new dimensions (W', H') are: W' = W * s H' = H * s 4. The image is resized to (W', H') using the specified interpolation method. 5. Bounding boxes and keypoints are scaled by the same factor s. Example: >>> import numpy as np >>> import albumentations as A >>> image = np.random.randint(0, 256, (100, 150, 3), dtype=np.uint8) >>> transform = A.SmallestMaxSize(max_size=120, p=1.0) >>> result = transform(image=image) >>> resized_image = result['image'] # resized_image will have shape (120, 180, 3), as the smallest side (100) # is scaled to 120, and the larger side is scaled proportionally c@s eZdZdS)zSmallestMaxSize.InitSchemaNr_r r r r!r)sr)r`Nrrar*r+r,rbcst||||_||_dSr>rcrdr3r r!r1 szSmallestMaxSize.__init__r;rrecKstj|||dSrf)r?Zsmallest_max_sizergr r r!r@szSmallestMaxSize.applyrEcKs|Sr>r rGr r r!rHszSmallestMaxSize.apply_to_bboxesrhcKs*|ddd}|t|}t|||Sri)minr?rJrmr r r!rK"s z"SmallestMaxSize.apply_to_keypointsrnr5cCs"dt|jtr|jn t|jiSrorpr9r r r!r:-szSmallestMaxSize.get_paramsrrcCsdSrfr r9r r r!rs0sz-SmallestMaxSize.get_transform_init_args_names)r$r%r&rMrrNrOrQrPrRrUr)rBrSr1r@rHrKr:rsrTr r r3r!rs+   cseZdZdZejejejejfZ Gddde Z e j ddfdddddd fd d Zd d d dddZd d d d dddZd d d dddZd d d dddZddddZZS)raHResize the input to the given height and width. Args: height (int): desired height of the output. width (int): desired width of the output. interpolation (OpenCV flag): flag that is used to specify the interpolation algorithm. Should be one of: cv2.INTER_NEAREST, cv2.INTER_LINEAR, cv2.INTER_CUBIC, cv2.INTER_AREA, cv2.INTER_LANCZOS4. Default: cv2.INTER_LINEAR. p (float): probability of applying the transform. Default: 1. Targets: image, mask, bboxes, keypoints Image types: uint8, float32 c@s:eZdZUeddZded<eddZded<ded<dS) zResize.InitSchemar)ger*heightwidthr rN)r$r%r&r rvr'rwr r r r!r)Is r)Nrr*r+r,)rvrwrr-r.cs&tj||d||_||_||_dSr/)r0r1rvrwr)r2rvrwrr-r.r3r r!r1NszResize.__init__r;r)r<r=rcKstj||j|jf|jdSN)r)r?resizervrwr)r2r<r=r r r!r@[sz Resize.apply)rAargsr=rcOstj||j|jftjdSrx)r?ryrvrwrBrC)r2rArzr=r r r!rD^szResize.apply_to_maskrEcKs|Sr>r rGr r r!rHaszResize.apply_to_bboxes)rIr=rcKs6|ddd\}}|j|}|j|}t|||Sri)rwrvr?rJ)r2rIr=rvrwZscale_xZscale_yr r r!rKes  zResize.apply_to_keypointsrrr5cCsdS)N)rvrwrr r9r r r!rsksz$Resize.get_transform_init_args_names)r$r%r&rMrrNrOrQrPrRr r)rBrSr1r@rDrHrKrsrTr r r3r!r4s  )% __future__rr7typingrrrrrrrBZnumpynpZpydanticr r r Zalbumentations.core.pydanticr Z(albumentations.core.transforms_interfacer rZalbumentations.core.typesrrrZalbumentations.core.utilsrrr?__all__rrUrrrr r r r!s      dR[