[fix] Improve dependencies for multi-backend support (#79)

* Add IMAGENET_DEFAULT_MEAN and IMAGENET_DEFAULT_STD

* Try update backend-specific tests

* modify test_effdet_backend ci

* add test_only_effdet.py

* more backend specific tests

* Add pytest

* escape torch.Tensor type annotation

* misc fixes

* install "torchvision" is more robust

Author: lolipopshock

.github/workflows/ci.yml

@@ -3,15 +3,68 @@ name: CI
 on: [push]
 
 jobs:
-  build:
 
+  test_only_effdet_backend:
+
+    runs-on: ubuntu-latest
+    steps:
+    - uses: actions/checkout@v2
+    - uses: actions/setup-python@v2
+      with:
+        python-version: '3.7'
+  
+    - name: Test Dependency Support 
+      run: |
+        pip install pytest
+        pip install -e . # The bare layoutparser module 
+        pytest tests_deps/test_file_utils.py
+
+    - name: Install only effdet deps 
+      run: |
+        pip install pytest
+        pip install -e ".[effdet]"
+        pytest tests_deps/test_only_effdet.py
+
+  test_only_detectron2_backend:
+
+    runs-on: ubuntu-latest
+    steps:
+    - uses: actions/checkout@v2
+    - uses: actions/setup-python@v2
+      with:
+        python-version: '3.7'
+
+    - name: Install only Detectron2 deps 
+      run: |
+        pip install pytest
+        pip install -e .
+        pip install torchvision && pip install "git+https://github.com/facebookresearch/detectron2.git@v0.5#egg=detectron2"
+        pytest tests_deps/test_only_detectron2.py
+
+  test_only_paddledetection_backend:
+
+    runs-on: ubuntu-latest
+    steps:
+    - uses: actions/checkout@v2
+    - uses: actions/setup-python@v2
+      with:
+        python-version: '3.7'
+
+    - name: Install only PaddleDetection deps 
+      run: |
+        pip install pytest
+        pip install -e ".[paddledetection]"
+        pytest tests_deps/test_only_paddledetection.py
+
+  test_all_methods_all_backends:
+    needs: [test_only_effdet_backend, test_only_detectron2_backend, test_only_paddledetection_backend]
     runs-on: ubuntu-latest
     strategy:
       matrix:
         python-version: [3.7, 3.8]
-      
     steps:
     - uses: actions/checkout@v2
+    
     - name: Set up Python ${{ matrix.python-version }}
       uses: actions/setup-python@v2
       with:
@@ -29,16 +82,10 @@ jobs:
         flake8 . --count --select=E9,F63,F7,F82 --show-source --statistics --ignore F821
         # exit-zero treats all errors as warnings. The GitHub editor is 127 chars wide
         flake8 . --count --exit-zero --max-complexity=10 --max-line-length=127 --statistics
-    
-    - name: Test Dependency Support 
-      run: |
-        # Install additional requirements when running tests
-        pip install pytest
-        pytest tests_deps
 
     - name: Test with pytest
       run: |
         # Install additional requirements when running tests
-        pip install .[effdet]
+        pip install ".[effdet]"
         pip install -r dev-requirements.txt
-        pytest tests
+        pytest tests

installation.md

@@ -15,7 +15,7 @@ After several major updates, LayoutParser provides various functionalities and d
 | --- | --- |
 | `pip install layoutparser`                   | **Install the base LayoutParser Library**<br>It will support all key functions in LayoutParser, including:<br />1. Layout Data Structure and operations<br />2. Layout Visualization <br />3. Load/export the layout data |
 | `pip install "layoutparser[effdet]"`           | **Install LayoutParser with Layout Detection Model Support**<br />It will install the LayoutParser base library as well as<br />supporting dependencies for the ***EfficientDet***-based layout detection models. |
-| `pip install layoutparser torch && pip install "git+https://github.com/facebookresearch/detectron2.git@v0.5#egg=detectron2"` | **Install LayoutParser with Layout Detection Model Support**<br />It will install the LayoutParser base library as well as<br />supporting dependencies for the ***Detectron2***-based layout detection models. See details in [Additional Instruction: Install Detectron2 Layout Model Backend](#additional-instruction-install-detectron2-layout-model-backend). |
+| `pip install layoutparser torchvision && pip install "git+https://github.com/facebookresearch/detectron2.git@v0.5#egg=detectron2"` | **Install LayoutParser with Layout Detection Model Support**<br />It will install the LayoutParser base library as well as<br />supporting dependencies for the ***Detectron2***-based layout detection models. See details in [Additional Instruction: Install Detectron2 Layout Model Backend](#additional-instruction-install-detectron2-layout-model-backend). |
 | `pip install "layoutparser[paddledetection]"`  | **Install LayoutParser with Layout Detection Model Support**<br />It will install the LayoutParser base library as well as<br />supporting dependencies for the ***PaddleDetection***-based layout detection models.  |
 | `pip install "layoutparser[ocr]"`              | **Install LayoutParser with OCR Support**<br />It will install the LayoutParser base library as well as<br />supporting dependencies for performing OCRs. See details in [Additional Instruction: Install OCR utils](#additional-instruction-install-ocr-utils).  |
 
@@ -26,7 +26,7 @@ After several major updates, LayoutParser provides various functionalities and d
 If you would like to use the Detectron2 models for layout detection, you might need to run the following command: 
 
 ```bash
-pip install layoutparser torch && pip install "detectron2@git+https://github.com/facebookresearch/detectron2.git@v0.5#egg=detectron2"
+pip install layoutparser torchvision && pip install "detectron2@git+https://github.com/facebookresearch/detectron2.git@v0.5#egg=detectron2"
 ```
 
 This might take some time as the command will *compile* the library. If you also want to install a Detectron2 version 

src/layoutparser/models/effdet/layoutmodel.py

@@ -31,6 +31,13 @@
         IMAGENET_DEFAULT_STD,
         transforms_coco_eval,
     )
+else:
+    # Copied from https://github.com/rwightman/efficientdet-pytorch/blob/c5b694aa34900fdee6653210d856ca8320bf7d4e/effdet/data/transforms.py#L13 
+    # Such that when effdet is not loaded, we'll still have default values for IMAGENET_DEFAULT_MEAN, IMAGENET_DEFAULT_STD
+    IMAGENET_DEFAULT_MEAN = (0.485, 0.456, 0.406)
+    IMAGENET_DEFAULT_STD = (0.229, 0.224, 0.225)
+    # IMAGENET_INCEPTION_MEAN = (0.5, 0.5, 0.5)
+    # IMAGENET_INCEPTION_STD = (0.5, 0.5, 0.5)
 
 
 class InputTransform:
@@ -56,7 +63,7 @@ def __init__(
         self.mean_tensor = torch.tensor([x * 255 for x in mean]).view(1, 3, 1, 1)
         self.std_tensor = torch.tensor([x * 255 for x in std]).view(1, 3, 1, 1)
 
-    def preprocess(self, image: Image) -> Tuple[torch.Tensor, Dict]:
+    def preprocess(self, image: Image) -> Tuple["torch.Tensor", Dict]:
 
         image = image.convert("RGB")
         image_info = {"img_size": image.size}
@@ -206,7 +213,7 @@ def detect(self, image: Union["np.ndarray", "Image.Image"]):
         layout = self.gather_output(model_outputs)
         return layout
 
-    def gather_output(self, model_outputs: torch.Tensor) -> Layout:
+    def gather_output(self, model_outputs: "torch.Tensor") -> Layout:
 
         model_outputs = model_outputs.cpu().detach()
         box_predictions = Layout()

tests_deps/test_only_detectron2.py

@@ -0,0 +1,31 @@
+# Copyright 2021 The Layout Parser team. 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 cv2
+import pytest
+from layoutparser import Detectron2LayoutModel
+
+def test_only_effdet_model():
+
+    # When all the backeds are not installed, it should 
+    # elicit only ImportErrors
+    
+    config = "lp://PubLayNet/faster_rcnn_R_50_FPN_3x/config"
+    model = Detectron2LayoutModel(config)
+    image = cv2.imread("tests/fixtures/model/test_model_image.jpg")
+    layout = model.detect(image)
+    
+    with pytest.raises(ImportError):
+        from layoutparser import EfficientDetLayoutModel
+        from layoutparser import PaddleDetectionLayoutModel

tests_deps/test_only_effdet.py

@@ -0,0 +1,31 @@
+# Copyright 2021 The Layout Parser team. 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 cv2
+import pytest
+from layoutparser import EfficientDetLayoutModel
+
+def test_only_effdet_model():
+
+    # When all the backeds are not installed, it should 
+    # elicit only ImportErrors
+    
+    config = "lp://PubLayNet/tf_efficientdet_d0/config"
+    model = EfficientDetLayoutModel(config)
+    image = cv2.imread("tests/fixtures/model/test_model_image.jpg")
+    layout = model.detect(image)
+    
+    with pytest.raises(ImportError):
+        from layoutparser import Detectron2LayoutModel
+        from layoutparser import PaddleDetectionLayoutModel

tests_deps/test_only_paddledetection.py

@@ -0,0 +1,31 @@
+# Copyright 2021 The Layout Parser team. 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 cv2
+import pytest
+from layoutparser import PaddleDetectionLayoutModel
+
+def test_only_effdet_model():
+
+    # When all the backeds are not installed, it should 
+    # elicit only ImportErrors
+    
+    config = "lp://PubLayNet/ppyolov2_r50vd_dcn_365e/config"
+    model = PaddleDetectionLayoutModel(config)
+    image = cv2.imread("tests/fixtures/model/test_model_image.jpg")
+    layout = model.detect(image)
+    
+    with pytest.raises(ImportError):
+        from layoutparser import EfficientDetLayoutModel
+        from layoutparser import Detectron2LayoutModel