[Feature] Support flip with label mapping

configs/recognition/tsm/README.md

@@ -46,7 +46,8 @@
 
 |config | resolution | gpus | backbone| pretrain | top1 acc (efficient/accurate)| top5 acc (efficient/accurate)| reference top1 acc (efficient/accurate)| reference top5 acc (efficient/accurate)| gpu_mem(M)  | ckpt | log| json|
 |:--|:--:|:--:|:--:|:--:|:--:|:--:|:--:|:--:|:--:|:--:|:--:|:--:|
-|[tsm_r50_1x1x8_50e_sthv1_rgb](/configs/recognition/tsm/tsm_r50_1x1x8_50e_sthv1_rgb.py) |height 100|8| ResNet50 | ImageNet|45.46 / 47.21|74.71 / 76.09|[45.50 / 47.33](https://github.com/mit-han-lab/temporal-shift-module/tree/8d53d6fda40bea2f1b37a6095279c4b454d672bd#training)|[74.34 / 76.60](https://github.com/mit-han-lab/temporal-shift-module/tree/8d53d6fda40bea2f1b37a6095279c4b454d672bd#training)| 7077| [ckpt](https://download.openmmlab.com/mmaction/recognition/tsm/tsm_r50_1x1x8_50e_sthv1_rgb/tsm_r50_1x1x8_50e_sthv1_rgb_20200616-3417f361.pth) | [log](https://download.openmmlab.com/mmaction/recognition/tsm/tsm_r50_1x1x8_50e_sthv1_rgb/20200616_022852.log)| [json](https://download.openmmlab.com/mmaction/recognition/tsm/tsm_r50_1x1x8_50e_sthv1_rgb/20200616_022852.log.json)|
+|[tsm_r50_1x1x8_50e_sthv1_rgb](/configs/recognition/tsm/tsm_r50_1x1x8_50e_sthv1_rgb.py) |height 100|8| ResNet50 | ImageNet| 45.58 / 47.70|75.02 / 76.12|[45.50 / 47.33](https://github.com/mit-han-lab/temporal-shift-module/tree/8d53d6fda40bea2f1b37a6095279c4b454d672bd#training)|[74.34 / 76.60](https://github.com/mit-han-lab/temporal-shift-module/tree/8d53d6fda40bea2f1b37a6095279c4b454d672bd#training)| 7077| [ckpt](https://download.openmmlab.com/mmaction/recognition/tsm/tsm_r50_1x1x8_50e_sthv1_rgb/tsm_r50_1x1x8_50e_sthv1_rgb_20210203-01dce462.pth) | [log](https://download.openmmlab.com/mmaction/recognition/tsm/tsm_r50_1x1x8_50e_sthv1_rgb/20210203_150227.log)| [json](https://download.openmmlab.com/mmaction/recognition/tsm/tsm_r50_1x1x8_50e_sthv1_rgb/20210203_150227.log.json)|
+|[tsm_r50_flip_1x1x8_50e_sthv1_rgb](/configs/recognition/tsm/tsm_r50_flip_1x1x8_50e_sthv1_rgb.py) |height 100|8| ResNet50 | ImageNet| 47.10 / 48.51|76.02 / 77.56|[45.50 / 47.33](https://github.com/mit-han-lab/temporal-shift-module/tree/8d53d6fda40bea2f1b37a6095279c4b454d672bd#training)|[74.34 / 76.60](https://github.com/mit-han-lab/temporal-shift-module/tree/8d53d6fda40bea2f1b37a6095279c4b454d672bd#training)| 7077| [ckpt](https://download.openmmlab.com/mmaction/recognition/tsm/tsm_r50_flip_1x1x8_50e_sthv1_rgb/tsm_r50_flip_1x1x8_50e_sthv1_rgb_20210203-12596f16.pth) | [log](https://download.openmmlab.com/mmaction/recognition/tsm/tsm_r50_flip_1x1x8_50e_sthv1_rgb/20210203_145829.log)| [json](https://download.openmmlab.com/mmaction/recognition/tsm/tsm_r50_flip_1x1x8_50e_sthv1_rgb/20210203_145829.log.json)|
 |[tsm_r50_1x1x16_50e_sthv1_rgb](/configs/recognition/tsm/tsm_r50_1x1x16_50e_sthv1_rgb.py)|height 100|8| ResNet50 | ImageNet|47.62 / 49.28|76.63 / 77.82|[47.05 / 48.61](https://github.com/mit-han-lab/temporal-shift-module/tree/8d53d6fda40bea2f1b37a6095279c4b454d672bd#training)|[76.40 / 77.96](https://github.com/mit-han-lab/temporal-shift-module/tree/8d53d6fda40bea2f1b37a6095279c4b454d672bd#training)|10390|[ckpt](https://download.openmmlab.com/mmaction/recognition/tsm/tsm_r50_1x1x16_50e_sthv1_rgb/tsm_r50_1x1x16_50e_sthv1_rgb_20201010-17fa49f6.pth)|[log](https://download.openmmlab.com/mmaction/recognition/tsm/tsm_r50_1x1x16_50e_sthv1_rgb/20201010_221240.log)|[json](https://download.openmmlab.com/mmaction/recognition/tsm/tsm_r50_1x1x16_50e_sthv1_rgb/20201010_221240.log.json)|
 |[tsm_r101_1x1x8_50e_sthv1_rgb](/configs/recognition/tsm/tsm_r101_1x1x8_50e_sthv1_rgb.py)|height 100|8| ResNet50 | ImageNet|45.72 / 48.43|74.67 / 76.72|[46.64 / 48.13](https://github.com/mit-han-lab/temporal-shift-module/tree/8d53d6fda40bea2f1b37a6095279c4b454d672bd#training)|[75.40 / 77.31](https://github.com/mit-han-lab/temporal-shift-module/tree/8d53d6fda40bea2f1b37a6095279c4b454d672bd#training)|9800|[ckpt](https://download.openmmlab.com/mmaction/recognition/tsm/tsm_r101_1x1x8_50e_sthv1_rgb/tsm_r101_1x1x8_50e_sthv1_rgb_20201010-43fedf2e.pth)|[log](https://download.openmmlab.com/mmaction/recognition/tsm/tsm_r101_1x1x8_50e_sthv1_rgb/20201010_224055.log)|[json](https://download.openmmlab.com/mmaction/recognition/tsm/tsm_r101_1x1x8_50e_sthv1_rgb/20201010_224055.log.json)|
 

configs/recognition/tsm/tsm_r50_flip_1x1x8_50e_sthv1_rgb.py

@@ -0,0 +1,97 @@
+_base_ = [
+    '../../_base_/models/tsm_r50.py', '../../_base_/schedules/sgd_tsm_50e.py',
+    '../../_base_/default_runtime.py'
+]
+
+# model settings
+model = dict(cls_head=dict(num_classes=174))
+
+# dataset settings
+dataset_type = 'RawframeDataset'
+data_root = 'data/sthv1/rawframes'
+data_root_val = 'data/sthv1/rawframes'
+ann_file_train = 'data/sthv1/sthv1_train_list_rawframes.txt'
+ann_file_val = 'data/sthv1/sthv1_val_list_rawframes.txt'
+ann_file_test = 'data/sthv1/sthv1_val_list_rawframes.txt'
+
+sthv1_flip_label_map = {2: 4, 4: 2, 30: 41, 41: 30, 52: 66, 66: 52}
+img_norm_cfg = dict(
+    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_bgr=False)
+
+train_pipeline = [
+    dict(type='SampleFrames', clip_len=1, frame_interval=1, num_clips=8),
+    dict(type='RawFrameDecode'),
+    dict(type='Resize', scale=(-1, 256)),
+    dict(
+        type='MultiScaleCrop',
+        input_size=224,
+        scales=(1, 0.875, 0.75, 0.66),
+        random_crop=False,
+        max_wh_scale_gap=1,
+        num_fixed_crops=13),
+    dict(type='Resize', scale=(224, 224), keep_ratio=False),
+    dict(type='Flip', flip_ratio=0.5, flip_label_map=sthv1_flip_label_map),
+    dict(type='Normalize', **img_norm_cfg),
+    dict(type='FormatShape', input_format='NCHW'),
+    dict(type='Collect', keys=['imgs', 'label'], meta_keys=[]),
+    dict(type='ToTensor', keys=['imgs', 'label'])
+]
+val_pipeline = [
+    dict(
+        type='SampleFrames',
+        clip_len=1,
+        frame_interval=1,
+        num_clips=8,
+        test_mode=True),
+    dict(type='RawFrameDecode'),
+    dict(type='Resize', scale=(-1, 256)),
+    dict(type='CenterCrop', crop_size=224),
+    dict(type='Normalize', **img_norm_cfg),
+    dict(type='FormatShape', input_format='NCHW'),
+    dict(type='Collect', keys=['imgs', 'label'], meta_keys=[]),
+    dict(type='ToTensor', keys=['imgs'])
+]
+test_pipeline = [
+    dict(
+        type='SampleFrames',
+        clip_len=1,
+        frame_interval=1,
+        num_clips=8,
+        test_mode=True),
+    dict(type='RawFrameDecode'),
+    dict(type='Resize', scale=(-1, 256)),
+    dict(type='CenterCrop', crop_size=224),
+    dict(type='Normalize', **img_norm_cfg),
+    dict(type='FormatShape', input_format='NCHW'),
+    dict(type='Collect', keys=['imgs', 'label'], meta_keys=[]),
+    dict(type='ToTensor', keys=['imgs'])
+]
+data = dict(
+    videos_per_gpu=8,
+    workers_per_gpu=4,
+    train=dict(
+        type=dataset_type,
+        ann_file=ann_file_train,
+        data_prefix=data_root,
+        filename_tmpl='{:05}.jpg',
+        pipeline=train_pipeline),
+    val=dict(
+        type=dataset_type,
+        ann_file=ann_file_val,
+        data_prefix=data_root_val,
+        filename_tmpl='{:05}.jpg',
+        pipeline=val_pipeline),
+    test=dict(
+        type=dataset_type,
+        ann_file=ann_file_test,
+        data_prefix=data_root_val,
+        filename_tmpl='{:05}.jpg',
+        pipeline=test_pipeline))
+evaluation = dict(
+    interval=2, metrics=['top_k_accuracy', 'mean_class_accuracy'])
+
+# optimizer
+optimizer = dict(weight_decay=0.0005)
+
+# runtime settings
+work_dir = './work_dirs/tsm_r50_flip_1x1x8_50e_sthv1_rgb/'

docs/changelog.md

@@ -9,6 +9,7 @@
 **New Features**
 
 - Support TSM-MobileNetV2 ([#415](https://github.com/open-mmlab/mmaction2/pull/415))
+- Support flip with label mapping ([#591](https://github.com/open-mmlab/mmaction2/pull/591))
 
 **ModelZoo**
 

docs/tutorials/4_data_pipeline.md

@@ -177,7 +177,7 @@ For each operation, we list the related dict fields that are added/updated/remov
 `Flip`
 
 - add: flip, flip_direction
-- update: imgs
+- update: imgs, label
 
 `Normalize`
 

mmaction/datasets/pipelines/augmentations.py

@@ -1091,16 +1091,23 @@ class Flip:
         flip_ratio (float): Probability of implementing flip. Default: 0.5.
         direction (str): Flip imgs horizontally or vertically. Options are
             "horizontal" | "vertical". Default: "horizontal".
+        flip_label_map (Dict[int, int] | None): Transform the label of the
+            flipped image with the specific label. Default: None.
         lazy (bool): Determine whether to apply lazy operation. Default: False.
     """
     _directions = ['horizontal', 'vertical']
 
-    def __init__(self, flip_ratio=0.5, direction='horizontal', lazy=False):
+    def __init__(self,
+                 flip_ratio=0.5,
+                 direction='horizontal',
+                 flip_label_map=None,
+                 lazy=False):
         if direction not in self._directions:
             raise ValueError(f'Direction {direction} is not supported. '
                              f'Currently support ones are {self._directions}')
         self.flip_ratio = flip_ratio
         self.direction = direction
+        self.flip_label_map = flip_label_map
         self.lazy = lazy
 
     def __call__(self, results):
@@ -1120,6 +1127,10 @@ def __call__(self, results):
         results['flip'] = flip
         results['flip_direction'] = self.direction
 
+        if self.flip_label_map is not None and flip:
+            results['label'] = self.flip_label_map.get(results['label'],
+                                                       results['label'])
+
         if not self.lazy:
             if flip:
                 for i, img in enumerate(results['imgs']):
@@ -1151,7 +1162,7 @@ def __repr__(self):
         repr_str = (
             f'{self.__class__.__name__}('
             f'flip_ratio={self.flip_ratio}, direction={self.direction}, '
-            f'lazy={self.lazy})')
+            f'flip_label_map={self.flip_label_map}, lazy={self.lazy})')
         return repr_str
 
 

tests/test_data/test_pipelines/test_augmentations/test_flip.py

@@ -78,4 +78,26 @@ def test_flip(self):
 
         assert repr(flip) == (f'{flip.__class__.__name__}'
                               f'(flip_ratio={1}, direction=vertical, '
-                              f'lazy={False})')
+                              f'flip_label_map={None}, lazy={False})')
+
+        # transform label for the flipped image with the specific label.
+        _flip_label_map = {4: 6}
+        imgs = list(np.random.rand(2, 64, 64, 3))
+
+        # the label should be mapped.
+        results = dict(imgs=copy.deepcopy(imgs), modality='RGB', label=4)
+        flip = Flip(
+            flip_ratio=1,
+            direction='horizontal',
+            flip_label_map=_flip_label_map)
+        flip_results = flip(results)
+        assert results['label'] == 6
+
+        # the label should not be mapped.
+        results = dict(imgs=copy.deepcopy(imgs), modality='RGB', label=3)
+        flip = Flip(
+            flip_ratio=1,
+            direction='horizontal',
+            flip_label_map=_flip_label_map)
+        flip_results = flip(results)
+        assert results['label'] == 3

tests/test_data/test_pipelines/test_augmentations/test_lazy.py

@@ -339,7 +339,7 @@ def test_flip_lazy(self):
 
         assert repr(flip) == (f'{flip.__class__.__name__}'
                               f'(flip_ratio={1}, direction=vertical, '
-                              f'lazy={True})')
+                              f'flip_label_map={None}, lazy={True})')
 
     def test_center_crop_lazy(self):
         with pytest.raises(TypeError):