[NOT Merge] add new model example pr

configs/my_example.yaml

@@ -0,0 +1,30 @@
+# PaddleVieo Example Configuration
+MODEL:
+    framework: "ExampleRecognizer"
+    backbone:
+        name: "ExampleBackbone"
+    head:
+        name: "ExampleHead"
+
+DATASET:
+    batch_size: 10
+    num_workers: 0
+    train:
+        format: "ExampleDataset"
+
+PIPELINE:
+    train:
+        norm:
+            name: "ExamplePipeline"  # NOTE: indent
+
+OPTIMIZER:
+    name: 'Adam'
+    learning_rate:
+        name: 'StepDecay'
+        learning_rate: 0.001
+        step_size: 10
+
+
+model_name: "ExampleModel"
+log_interval: 1
+epochs: 5

paddlevideo/loader/dataset/__init__.py

@@ -18,8 +18,9 @@
 from .bmn_dataset import BMNDataset
 from .feature import FeatureDataset
 from .skeleton import SkeletonDataset
+from .example_dataset import ExampleDataset
 
 __all__ = [
     'VideoDataset', 'FrameDataset', 'SFVideoDataset', 'BMNDataset',
-    'FeatureDataset', 'SkeletonDataset'
+    'FeatureDataset', 'SkeletonDataset', 'ExampleDataset'
 ]

paddlevideo/loader/dataset/example_dataset.py

@@ -0,0 +1,41 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import copy
+import numpy as np
+
+from ..registry import DATASETS
+from .base import BaseDataset
+
+
+@DATASETS.register()
+class ExampleDataset(BaseDataset):
+    """ ExampleDataset """
+    def __init__(self, pipeline, file_path=''):
+        super().__init__(file_path, pipeline)
+
+    def load_file(self):
+        """load file, abstractmethod"""
+        self.x = np.random.rand(100, 20, 20)
+        self.y = np.random.randint(10, size=(100, 1))
+
+    def prepare_train(self, idx):
+        """Prepare the frames for training/valid given index. """
+        results = {'data': self.x[idx], 'label': self.y[idx]}
+        results = self.pipeline(results)
+        return results['data'], results['label']
+
+    def __len__(self):
+        """get the size of the dataset."""
+        return self.x.shape[0]

paddlevideo/loader/pipelines/__init__.py

@@ -22,11 +22,12 @@
 from .mix import Cutmix, Mixup
 from .sample import Sampler
 from .skeleton_pipeline import AutoPadding, Iden, SkeletonNorm
+from .example_pipeline import ExamplePipeline
 
 __all__ = [
     'Scale', 'RandomCrop', 'CenterCrop', 'RandomFlip', 'Image2Array',
     'Normalization', 'Compose', 'VideoDecoder', 'FrameDecoder', 'Sampler',
     'Mixup', 'Cutmix', 'JitterScale', 'MultiCrop', 'PackOutput', 'TenCrop',
     'UniformCrop', 'DecodeSampler', 'LoadFeat', 'GetMatchMap', 'GetVideoLabel',
-    'AutoPadding', 'SkeletonNorm', 'Iden'
+    'AutoPadding', 'SkeletonNorm', 'Iden', 'ExamplePipeline'
 ]

paddlevideo/loader/pipelines/example_pipeline.py

@@ -0,0 +1,31 @@
+#  Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from ..registry import PIPELINES
+
+
+@PIPELINES.register()
+class ExamplePipeline(object):
+    """Example Pipeline """
+    def __init__(self, mean=0, std=1.0):
+        self.mean = mean
+        self.std = std
+
+    def __call__(self, results):
+        data = results['data']
+        label = results['label']
+        norm_data = (data - self.mean) / self.std
+        results['data'] = norm_data.astype('float32')
+        results['label'] = label.astype('int64')
+        return results

paddlevideo/modeling/backbones/__init__.py

@@ -21,8 +21,9 @@
 from .vit import VisionTransformer
 from .stgcn import STGCN
 from .agcn import AGCN
+from .example_backbone import ExampleBackbone
 
 __all__ = [
     'ResNet', 'ResNetTSM', 'ResNetTweaksTSM', 'ResNetSlowFast', 'BMN',
-    'ResNetTweaksTSN', 'VisionTransformer', 'STGCN', 'AGCN'
+    'ResNetTweaksTSN', 'VisionTransformer', 'STGCN', 'AGCN', 'ExampleBackbone'
 ]

paddlevideo/modeling/backbones/example_backbone.py

@@ -0,0 +1,39 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import paddle
+import paddle.nn as nn
+import paddle.nn.functional as F
+
+from ..registry import BACKBONES
+
+
+@BACKBONES.register()
+class ExampleBackbone(nn.Layer):
+    """Example Backbone"""
+    def __init__(self):
+        super(ExampleBackbone, self).__init__()
+        ## 2.1 网络Backbobe
+        self.layer1 = paddle.nn.Flatten(1, -1)
+        self.layer2 = paddle.nn.Linear(400, 512)
+        self.layer3 = paddle.nn.ReLU()
+        self.layer4 = paddle.nn.Dropout(0.2)
+
+    def forward(self, x):
+        """ model forward"""
+        y = self.layer1(x)
+        y = self.layer2(y)
+        y = self.layer3(y)
+        y = self.layer4(y)
+        return y

paddlevideo/modeling/framework/recognizers/__init__.py

@@ -16,8 +16,9 @@
 from .recognizer3d import Recognizer3D
 from .recognizer_transformer import RecognizerTransformer
 from .recognizer_gcn import RecognizerGCN
+from .example_recognizer import ExampleRecognizer
 
 __all__ = [
     'BaseRecognizer', 'Recognizer1D', 'Recognizer2D', 'Recognizer3D',
-    'RecognizerTransformer', 'RecognizerGCN'
+    'RecognizerTransformer', 'RecognizerGCN', 'ExampleRecognizer'
 ]

paddlevideo/modeling/framework/recognizers/example_recognizer.py

@@ -0,0 +1,34 @@
+# Copyright (c) 2021  PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License"
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+
+from ...registry import RECOGNIZERS
+from .base import BaseRecognizer
+import paddle
+
+
+@RECOGNIZERS.register()
+class ExampleRecognizer(BaseRecognizer):
+    """Example Recognizer model framework."""
+    def forward_net(self, imgs):
+        """model forward method"""
+        feature = self.backbone(imgs)
+        cls_score = self.head(feature)
+        return cls_score
+
+    def train_step(self, data_batch):
+        """Define how the model is going to train, from input to output.
+        """
+        imgs = data_batch[0]
+        labels = data_batch[1:]
+        cls_score = self.forward_net(imgs)
+        loss_metrics = self.head.loss(cls_score, labels)
+        return loss_metrics

paddlevideo/modeling/heads/__init__.py

@@ -21,8 +21,9 @@
 from .attention_lstm_head import AttentionLstmHead
 from .timesformer_head import TimeSformerHead
 from .stgcn_head import STGCNHead
+from .example_head import ExampleHead
 
 __all__ = [
     'BaseHead', 'TSNHead', 'TSMHead', 'ppTSMHead', 'ppTSNHead', 'SlowFastHead',
-    'AttentionLstmHead', 'TimeSformerHead', 'STGCNHead'
+    'AttentionLstmHead', 'TimeSformerHead', 'STGCNHead', 'ExampleHead'
 ]

paddlevideo/modeling/heads/example_head.py

@@ -0,0 +1,31 @@
+# Copyright (c) 2021  PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License"
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import paddle
+import paddle.nn as nn
+from ..registry import HEADS
+from .base import BaseHead
+
+
+@HEADS.register()
+class ExampleHead(BaseHead):
+    """ Example Head """
+    def __init__(self, num_classes=10, in_channels=512):
+        super().__init__(num_classes, in_channels)
+        self.head = paddle.nn.Linear(in_channels, num_classes)
+
+    def forward(self, x):
+        """model forward """
+        y = self.head(x)
+        return y

run.sh

@@ -7,6 +7,9 @@ export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
 
 start_time=$(date +%s)
 
+# run example model
+python3.7 main.py -c configs/my_example.yaml
+
 # run tsm training
 #python3.7 -B -m paddle.distributed.launch --gpus="0,1,2,3,4,5,6,7" --log_dir=log_tsm main.py  --validate -c configs/recognition/tsm/tsm_k400_frames.yaml
 
@@ -32,7 +35,7 @@ start_time=$(date +%s)
 #python3.7 -B -m paddle.distributed.launch --gpus="0,1,2,3,4,5,6,7" --log_dir=log_attetion_lstm  main.py  --validate -c configs/recognition/attention_lstm/attention_lstm_youtube-8m.yaml
 
 # run pp-tsm training
-python3.7 -B -m paddle.distributed.launch --gpus="0,1,2,3,4,5,6,7"  --log_dir=log_pptsm  main.py  --validate -c configs/recognition/pptsm/pptsm_k400_frames_uniform.yaml
+#python3.7 -B -m paddle.distributed.launch --gpus="0,1,2,3,4,5,6,7"  --log_dir=log_pptsm  main.py  --validate -c configs/recognition/pptsm/pptsm_k400_frames_uniform.yaml
 
 # run pp-tsn training
 # python3.7 -B -m paddle.distributed.launch --gpus="0,1,2,3,4,5,6,7"  --log_dir=log_pptsn  main.py  --validate -c configs/recognition/pptsn/pptsn_k400_frames.yaml