Add Shear Augmentation.

mmdet/core/mask/structures.py

@@ -1,5 +1,6 @@
 from abc import ABCMeta, abstractmethod
 
+import cv2
 import mmcv
 import numpy as np
 import pycocotools.mask as maskUtils
@@ -132,6 +133,36 @@ def to_tensor(self, dtype, device):
         """
         pass
 
+    @staticmethod
+    def warpAffine(data,
+                   trans_matrix,
+                   out_size,
+                   fill_val,
+                   flags=cv2.INTER_NEAREST,
+                   borderMode=cv2.BORDER_CONSTANT):
+        """Affine wrapper which transforms the source data using the given
+        trans_matrix.
+
+        Args:
+            data (np.ndarray): Source data.
+            trans_matrix (np.ndarray): Transformation matrix with shape (2, 3).
+            out_size (tuple): Size of the output data with format (w, h).
+            fill_val (int | float | tuple): Value used in case of a constant
+                border.
+            flags: Interpolation methods used in ``cv2.warpAffine``.
+            borderMode: pixel extrapolation method used in ``cv2.warpAffine``.
+
+        Returns:
+            np.ndarray: transformed data with the same shape as input data.
+        """
+        return cv2.warpAffine(
+            data,
+            trans_matrix,
+            dsize=out_size,  # dsize takes input size as order (w,h).
+            flags=flags,
+            borderMode=borderMode,
+            borderValue=fill_val)
+
 
 class BitmapMasks(BaseInstanceMasks):
     """This class represents masks in the form of bitmaps.
@@ -297,6 +328,21 @@ def expand(self, expanded_h, expanded_w, top, left):
                           left:left + self.width] = self.masks
         return BitmapMasks(expanded_mask, expanded_h, expanded_w)
 
+    def shear(self,
+              shear_matrix,
+              out_shape,
+              fill_val=0,
+              flags=cv2.INTER_NEAREST,
+              borderMode=cv2.BORDER_CONSTANT):
+        if len(self.masks) == 0:
+            sheared_masks = np.empty((0, *out_shape), dtype=np.uint8)
+        else:
+            sheared_masks = np.stack([
+                self.warpAffine(mask, shear_matrix, out_shape[::-1], fill_val,
+                                flags, borderMode) for mask in self.masks
+            ]).astype(self.masks.dtype)
+        return BitmapMasks(sheared_masks, *out_shape)
+
     @property
     def areas(self):
         """See :py:attr:`BaseInstanceMasks.areas`."""
@@ -498,6 +544,14 @@ def crop_and_resize(self,
             resized_masks.append(resized_mask)
         return PolygonMasks(resized_masks, *out_shape)
 
+    def shear(self,
+              shear_matrix,
+              out_shape,
+              fill_val=0,
+              flags=cv2.INTER_NEAREST,
+              borderMode=cv2.BORDER_CONSTANT):
+        raise NotImplementedError
+
     def to_bitmap(self):
         """convert polygon masks to bitmap masks."""
         bitmap_masks = self.to_ndarray()

mmdet/datasets/pipelines/__init__.py

@@ -5,16 +5,40 @@
 from .instaboost import InstaBoost
 from .loading import (LoadAnnotations, LoadImageFromFile,
                       LoadMultiChannelImageFromFiles, LoadProposals)
+from .shear import Shear
 from .test_time_aug import MultiScaleFlipAug
 from .transforms import (Albu, CutOut, Expand, MinIoURandomCrop, Normalize,
                          Pad, PhotoMetricDistortion, RandomCenterCropPad,
                          RandomCrop, RandomFlip, Resize, SegRescale)
 
 __all__ = [
-    'Compose', 'to_tensor', 'ToTensor', 'ImageToTensor', 'ToDataContainer',
-    'Transpose', 'Collect', 'LoadAnnotations', 'LoadImageFromFile',
-    'LoadMultiChannelImageFromFiles', 'LoadProposals', 'MultiScaleFlipAug',
-    'Resize', 'RandomFlip', 'Pad', 'RandomCrop', 'Normalize', 'SegRescale',
-    'MinIoURandomCrop', 'Expand', 'PhotoMetricDistortion', 'Albu',
-    'InstaBoost', 'RandomCenterCropPad', 'AutoAugment', 'CutOut'
+    'Compose',
+    'to_tensor',
+    'ToTensor',
+    'ImageToTensor',
+    'ToDataContainer',
+    'Transpose',
+    'Collect',
+    'LoadAnnotations',
+    'LoadImageFromFile',
+    'LoadMultiChannelImageFromFiles',
+    'LoadProposals',
+    'MultiScaleFlipAug',
+    'Resize',
+    'RandomFlip',
+    'Pad',
+    'RandomCrop',
+    'Normalize',
+    'SegRescale',
+    'MinIoURandomCrop',
+    'Expand',
+    'PhotoMetricDistortion',
+    'Albu',
+    'InstaBoost',
+    'RandomCenterCropPad',
+    'AutoAugment',
+    'CutOut',
+    'Translate',
+    'TranslateOnlyBBox',
+    'Shear',
 ]

mmdet/datasets/pipelines/auto_augment.py

@@ -23,23 +23,23 @@ class AutoAugment(object):
             augment images.
 
     Examples:
-        >>> replace = (104, 116, 124)
+        >>> fill_val = (104, 116, 124)
         >>> policies = [
         >>>     [
         >>>         dict(type='Sharpness', prob=0.0, level=8),
         >>>         dict(
         >>>             type='Shear',
         >>>             prob=0.4,
         >>>             level=0,
-        >>>             replace=replace,
+        >>>             fill_val=fill_val,
         >>>             axis='x')
         >>>     ],
         >>>     [
         >>>         dict(
         >>>             type='Rotate',
         >>>             prob=0.6,
         >>>             level=10,
-        >>>             replace=replace),
+        >>>             fill_val=fill_val),
         >>>         dict(type='Color', prob=1.0, level=6)
         >>>     ]
         >>> ]
@@ -65,8 +65,17 @@ def __init__(self, policies):
         self.transforms = [Compose(policy) for policy in self.policies]
 
     def __call__(self, results):
+        """Call function to randomly select transforms and perform
+        transformation.
+
+        Args:
+            results (dict): Result dict from loading pipeline.
+
+        Returns:
+            dict: Transformed results.
+        """
         transform = np.random.choice(self.transforms)
         return transform(results)
 
     def __repr__(self):
-        return f'{self.__class__.__name__}(policies={self.policies}'
+        return f'{self.__class__.__name__}(policies={self.policies})'