Merge from master

.circleci/config.yml

@@ -25,6 +25,9 @@ jobs:
           name: Run Pytorch scripts
           command: ./scripts/run_pytorch.sh
           no_output_timeout: 1h
+      - store_test_results:
+          path: test-results
+
 
 workflows:
   version: 2

pytorch_vision_alexnet.md

@@ -8,7 +8,7 @@ category: researchers
 image: alexnet2.png
 author: Pytorch Team
 tags: [vision, scriptable]
-github-link: https://github.com/pytorch/vision/blob/master/torchvision/models/alexnet.py
+github-link: https://github.com/pytorch/vision/blob/main/torchvision/models/alexnet.py
 github-id: pytorch/vision
 featured_image_1: alexnet1.png
 featured_image_2: alexnet2.png

pytorch_vision_deeplabv3_resnet101.md

@@ -2,13 +2,13 @@
 layout: hub_detail
 background-class: hub-background
 body-class: hub
-title: Deeplabv3-ResNet101
-summary: DeepLabV3 model with a ResNet-101 backbone
+title: Deeplabv3
+summary: DeepLabV3 models with ResNet-50, ResNet-101 and MobileNet-V3 backbones
 category: researchers
 image: deeplab2.png
 author: Pytorch Team
 tags: [vision, scriptable]
-github-link: https://github.com/pytorch/vision/blob/master/torchvision/models/segmentation/deeplabv3.py
+github-link: https://github.com/pytorch/vision/blob/main/torchvision/models/segmentation/deeplabv3.py
 github-id: pytorch/vision
 featured_image_1: deeplab1.png
 featured_image_2: deeplab2.png
@@ -18,7 +18,10 @@ order: 1
 
 ```python
 import torch
-model = torch.hub.load('pytorch/vision:v0.10.0', 'deeplabv3_resnet101', pretrained=True)
+model = torch.hub.load('pytorch/vision:v0.10.0', 'deeplabv3_resnet50', pretrained=True)
+# or any of these variants
+# model = torch.hub.load('pytorch/vision:v0.10.0', 'deeplabv3_resnet101', pretrained=True)
+# model = torch.hub.load('pytorch/vision:v0.10.0', 'deeplabv3_mobilenet_v3_large', pretrained=True)
 model.eval()
 ```
 
@@ -35,7 +38,7 @@ So, `output['out']` is of shape `(N, 21, H, W)`. More documentation can be found
 ```python
 # Download an example image from the pytorch website
 import urllib
-url, filename = ("https://github.com/pytorch/hub/raw/master/images/dog.jpg", "dog.jpg")
+url, filename = ("https://github.com/pytorch/hub/raw/master/images/deeplab1.png", "deeplab1.png")
 try: urllib.URLopener().retrieve(url, filename)
 except: urllib.request.urlretrieve(url, filename)
 ```
@@ -45,6 +48,7 @@ except: urllib.request.urlretrieve(url, filename)
 from PIL import Image
 from torchvision import transforms
 input_image = Image.open(filename)
+input_image = input_image.convert("RGB")
 preprocess = transforms.Compose([
     transforms.ToTensor(),
     transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
@@ -86,14 +90,17 @@ plt.imshow(r)
 
 ### Model Description
 
-Deeplabv3-ResNet101 is constructed by a Deeplabv3 model with a ResNet-101 backbone.
+Deeplabv3-ResNet is constructed by a Deeplabv3 model using a ResNet-50 or ResNet-101 backbone.
+Deeplabv3-MobileNetV3-Large is constructed by a Deeplabv3 model using the MobileNetV3 large backbone.
 The pre-trained model has been trained on a subset of COCO train2017, on the 20 categories that are present in the Pascal VOC dataset.
 
 Their accuracies of the pre-trained models evaluated on COCO val2017 dataset are listed below.
 
-|    Model structure  |   Mean IOU  | Global Pixelwise Accuracy |
-| ------------------- | ----------- | --------------------------|
-| deeplabv3_resnet101 |   67.4      |   92.4                    |
+|    Model structure           |   Mean IOU  | Global Pixelwise Accuracy |
+| ---------------------------- | ----------- | --------------------------|
+| deeplabv3_resnet50           |   66.4      |   92.4                    |
+| deeplabv3_resnet101          |   67.4      |   92.4                    |
+| deeplabv3_mobilenet_v3_large |   60.3      |   91.2                    |
 
 ### Resources
 

pytorch_vision_densenet.md

@@ -8,7 +8,7 @@ category: researchers
 image: densenet1.png
 author: Pytorch Team
 tags: [vision, scriptable]
-github-link: https://github.com/pytorch/vision/blob/master/torchvision/models/densenet.py
+github-link: https://github.com/pytorch/vision/blob/main/torchvision/models/densenet.py
 github-id: pytorch/vision
 featured_image_1: densenet1.png
 featured_image_2: densenet2.png

pytorch_vision_fcn_resnet101.md

@@ -2,13 +2,13 @@
 layout: hub_detail
 background-class: hub-background
 body-class: hub
-title: FCN-ResNet101
-summary: Fully-Convolutional Network model with a ResNet-101 backbone
+title: FCN
+summary: Fully-Convolutional Network model with ResNet-50 and ResNet-101 backbones
 category: researchers
 image: fcn2.png
 author: Pytorch Team
 tags: [vision, scriptable]
-github-link: https://github.com/pytorch/vision/blob/master/torchvision/models/segmentation/fcn.py
+github-link: https://github.com/pytorch/vision/blob/main/torchvision/models/segmentation/fcn.py
 github-id: pytorch/vision
 featured_image_1: deeplab1.png
 featured_image_2: fcn2.png
@@ -18,7 +18,9 @@ order: 10
 
 ```python
 import torch
-model = torch.hub.load('pytorch/vision:v0.10.0', 'fcn_resnet101', pretrained=True)
+model = torch.hub.load('pytorch/vision:v0.10.0', 'fcn_resnet50', pretrained=True)
+# or
+# model = torch.hub.load('pytorch/vision:v0.10.0', 'fcn_resnet101', pretrained=True)
 model.eval()
 ```
 
@@ -35,7 +37,7 @@ So, `output['out']` is of shape `(N, 21, H, W)`. More documentation can be found
 ```python
 # Download an example image from the pytorch website
 import urllib
-url, filename = ("https://github.com/pytorch/hub/raw/master/images/dog.jpg", "dog.jpg")
+url, filename = ("https://github.com/pytorch/hub/raw/master/images/deeplab1.png", "deeplab1.png")
 try: urllib.URLopener().retrieve(url, filename)
 except: urllib.request.urlretrieve(url, filename)
 ```
@@ -45,6 +47,7 @@ except: urllib.request.urlretrieve(url, filename)
 from PIL import Image
 from torchvision import transforms
 input_image = Image.open(filename)
+input_image = input_image.convert("RGB")
 preprocess = transforms.Compose([
     transforms.ToTensor(),
     transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
@@ -85,13 +88,14 @@ plt.imshow(r)
 
 ### Model Description
 
-FCN-ResNet101 is constructed by a Fully-Convolutional Network model with a ResNet-101 backbone.
+FCN-ResNet is constructed by a Fully-Convolutional Network model, using a ResNet-50 or a ResNet-101 backbone.
 The pre-trained models have been trained on a subset of COCO train2017, on the 20 categories that are present in the Pascal VOC dataset.
 
 Their accuracies of the pre-trained models evaluated on COCO val2017 dataset are listed below.
 
 | Model structure |   Mean IOU  | Global Pixelwise Accuracy |
 | --------------- | ----------- | --------------------------|
+|  fcn_resnet50   |   60.5      |   91.4                    |
 |  fcn_resnet101  |   63.7      |   91.9                    |
 
 ### Resources

pytorch_vision_googlenet.md

@@ -8,7 +8,7 @@ category: researchers
 image: googlenet1.png
 author: Pytorch Team
 tags: [vision]
-github-link: https://github.com/pytorch/vision/blob/master/torchvision/models/googlenet.py
+github-link: https://github.com/pytorch/vision/blob/main/torchvision/models/googlenet.py
 github-id: pytorch/vision
 featured_image_1: googlenet1.png
 featured_image_2: googlenet2.png

pytorch_vision_inception_v3.md

@@ -8,7 +8,7 @@ category: researchers
 image: inception_v3.png
 author: Pytorch Team
 tags: [vision, scriptable]
-github-link: https://github.com/pytorch/vision/blob/master/torchvision/models/inception.py
+github-link: https://github.com/pytorch/vision/blob/main/torchvision/models/inception.py
 github-id: pytorch/vision
 featured_image_1: inception_v3.png
 featured_image_2: no-image

pytorch_vision_mobilenet_v2.md

@@ -8,7 +8,7 @@ category: researchers
 image: mobilenet_v2_1.png
 author: Pytorch Team
 tags: [vision, scriptable]
-github-link: https://github.com/pytorch/vision/blob/master/torchvision/models/mobilenet.py
+github-link: https://github.com/pytorch/vision/blob/main/torchvision/models/mobilenet.py
 github-id: pytorch/vision
 featured_image_1: mobilenet_v2_1.png
 featured_image_2: mobilenet_v2_2.png

pytorch_vision_resnet.md

@@ -8,7 +8,7 @@ category: researchers
 image: resnet.png
 author: Pytorch Team
 tags: [vision, scriptable]
-github-link: https://github.com/pytorch/vision/blob/master/torchvision/models/resnet.py
+github-link: https://github.com/pytorch/vision/blob/main/torchvision/models/resnet.py
 github-id: pytorch/vision
 featured_image_1: resnet.png
 featured_image_2: no-image

pytorch_vision_resnext.md

@@ -8,7 +8,7 @@ category: researchers
 image: resnext.png
 author: Pytorch Team
 tags: [vision, scriptable]
-github-link: https://github.com/pytorch/vision/blob/master/torchvision/models/resnet.py
+github-link: https://github.com/pytorch/vision/blob/main/torchvision/models/resnet.py
 github-id: pytorch/vision
 featured_image_1: resnext.png
 featured_image_2: no-image

pytorch_vision_shufflenet_v2.md

@@ -8,7 +8,7 @@ category: researchers
 image: shufflenet_v2_1.png
 author: Pytorch Team
 tags: [vision, scriptable]
-github-link: https://github.com/pytorch/vision/blob/master/torchvision/models/shufflenetv2.py
+github-link: https://github.com/pytorch/vision/blob/main/torchvision/models/shufflenetv2.py
 github-id: pytorch/vision
 featured_image_1: shufflenet_v2_1.png
 featured_image_2: shufflenet_v2_2.png

pytorch_vision_squeezenet.md

@@ -8,7 +8,7 @@ category: researchers
 image: squeezenet.png
 author: Pytorch Team
 tags: [vision, scriptable]
-github-link: https://github.com/pytorch/vision/blob/master/torchvision/models/squeezenet.py
+github-link: https://github.com/pytorch/vision/blob/main/torchvision/models/squeezenet.py
 github-id: pytorch/vision
 featured_image_1: squeezenet.png
 featured_image_2: no-image

pytorch_vision_vgg.md

@@ -8,7 +8,7 @@ category: researchers
 image: vgg.png
 author: Pytorch Team
 tags: [vision, scriptable]
-github-link: https://github.com/pytorch/vision/blob/master/torchvision/models/vgg.py
+github-link: https://github.com/pytorch/vision/blob/main/torchvision/models/vgg.py
 github-id: pytorch/vision
 featured_image_1: vgg.png
 featured_image_2: no-image

pytorch_vision_wide_resnet.md

@@ -8,7 +8,7 @@ category: researchers
 image: wide_resnet.png
 author: Sergey Zagoruyko
 tags: [vision, scriptable]
-github-link: https://github.com/pytorch/vision/blob/master/torchvision/models/resnet.py
+github-link: https://github.com/pytorch/vision/blob/main/torchvision/models/resnet.py
 github-id: pytorch/vision
 featured_image_1: wide_resnet.png
 featured_image_2: no-image

scripts/install.sh

@@ -6,6 +6,8 @@ conda activate base
 
 conda install -y pytorch torchvision torchaudio -c pytorch-nightly
 
+conda install -y pytest
+
 # Dependencies required to load models
 conda install -y regex pillow tqdm boto3 requests numpy\
     h5py scipy matplotlib unidecode ipython pyyaml

scripts/run_pytorch.sh

@@ -2,40 +2,20 @@
 . ~/miniconda3/etc/profile.d/conda.sh
 conda activate base
 
-ALL_FILE=$(find *.md ! -name README.md)
-TEMP_PY="temp.py"
-CUDAS="nvidia"
+ALL_FILES=$(find *.md ! -name README.md)
+PYTHON_CODE_DIR="python_code"
 
-declare -i error_code=0
+mkdir $PYTHON_CODE_DIR
 
-for f in $ALL_FILE
+# Quick rundown: for each file we extract the python code that's within
+# the ``` markers and we put that code in a corresponding .py file in $PYTHON_CODE_DIR
+# Then we execute each of these python files with pytest in test_run_python_code.py
+for f in $ALL_FILES
 do
-  echo "Running pytorch example in $f"
-  # FIXME: NVIDIA models checkoints are on cuda
-  if [[ $f = $CUDAS* ]]; then
-    echo "...skipped due to cuda checkpoints."
-  elif [[ $f = "pytorch_fairseq_translation"* ]]; then
-    echo "...temporarily disabled"
-  # FIXME: torch.nn.modules.module.ModuleAttributeError: 'autoShape' object has no attribute 'fuse'
-  elif [[ $f = "ultralytics_yolov5"* ]]; then
-    echo "...temporarily disabled"
-  elif [[ $f = "huggingface_pytorch-transformers"* ]]; then
-    echo "...temporarily disabled"
-  elif [[ $f = "peterl1n_robustvideomatting"* ]]; then
-    echo "...temporarily disabled"
-  # FIXME: TypeError: compose() got an unexpected keyword argument 'strict'
-  elif [[ $f = "pytorch_fairseq_roberta"* ]]; then
-    echo "...temporarily disabled"
-  # FIXME: rate limiting
-  else
-    sed -n '/^```python/,/^```/ p' < $f | sed '/^```/ d' > $TEMP_PY
-    python $TEMP_PY
-    error_code+=$?
-
-    if [ -f "$TEMP_PY" ]; then
-      rm $TEMP_PY
-    fi
-  fi
+  f_no_ext=${f%.md}  # remove .md extension
+  out_py=$PYTHON_CODE_DIR/$f_no_ext.py
+  echo "Extracting Python code from $f into $out_py"
+  sed -n '/^```python/,/^```/ p' < $f | sed '/^```/ d' > $out_py
 done
 
-exit $error_code
+pytest --junitxml=test-results/junit.xml test_run_python_code.py -vv

test_run_python_code.py

@@ -0,0 +1,43 @@
+from subprocess import check_output, STDOUT, CalledProcessError
+import sys
+import pytest
+import glob
+
+
+PYTHON_CODE_DIR = "python_code"
+ALL_FILES = glob.glob(PYTHON_CODE_DIR + "/*.py")
+
+
+@pytest.mark.parametrize('file_path', ALL_FILES)
+def test_run_file(file_path):
+    if 'nvidia' in file_path:
+        # FIXME: NVIDIA models checkoints are on cuda
+        pytest.skip("temporarily disabled")
+    if 'pytorch_fairseq_translation' in file_path:
+        pytest.skip("temporarily disabled")
+    if 'ultralytics_yolov5' in file_path:
+        # FIXME torch.nn.modules.module.ModuleAttributeError: 'autoShape' object has no attribute 'fuse
+        pytest.skip("temporarily disabled")
+    if 'huggingface_pytorch-transformers' in file_path:
+        # FIXME torch.nn.modules.module.ModuleAttributeError: 'autoShape' object has no attribute 'fuse
+        pytest.skip("temporarily disabled")
+    if 'pytorch_fairseq_roberta' in file_path:
+        pytest.skip("temporarily disabled")
+    if 'robustvideomatting' in file_path:
+        pytest.skip("temporarily disabled")
+
+    # We just run the python files in a separate sub-process. We really want a
+    # subprocess here because otherwise we might run into package versions
+    # issues: imagine script A that needs torchvivion 0.9 and script B that
+    # needs torchvision 0.10. If script A is run prior to script B in the same
+    # process, script B will still be run with torchvision 0.9 because the only
+    # "import torchvision" statement that counts is the first one, and even
+    # torchub sys.path shenanigans can do nothing about this. By creating
+    # subprocesses we're sure that all file executions are fully independent.
+    try:
+        # This is inspired (and heavily simplified) from
+        # https://github.com/cloudpipe/cloudpickle/blob/343da119685f622da2d1658ef7b3e2516a01817f/tests/testutils.py#L177
+        out = check_output([sys.executable, file_path], stderr=STDOUT)
+        print(out.decode())
+    except CalledProcessError as e:
+        raise RuntimeError(f"Script {file_path} errored with output:\n{e.output.decode()}")