update CI

.circleci/config.yml

@@ -10,31 +10,55 @@ cpu: &cpu
 
 gpu: &gpu
   machine:
-    image: ubuntu-1604:201903-01
-    docker_layer_caching: true
+    image: ubuntu-1604-cuda-10.1:201909-23
   resource_class: gpu.small
 
 windows-cpu: &windows_cpu
-    machine:
-      resource_class: windows.medium
-      image: windows-server-2019-vs2019:stable
-      shell: powershell.exe
+  machine:
+    resource_class: windows.medium
+    image: windows-server-2019-vs2019:stable
+    shell: powershell.exe
 
 # windows-gpu: &windows_gpu
 #     machine:
 #       resource_class: windows.gpu.nvidia.medium
 #       image: windows-server-2019-nvidia:stable
 
+pytorch_parameters: &pytorch_parameters
+  parameters:
+    pytorch_version:
+      type: string
+      default: "1.6"
+    torchvision_version:
+      type: string
+      default: "0.7"
+
+  environment:
+    PYTORCH_VERSION: << parameters.pytorch_version >>
+    TORCHVISION_VERSION: << parameters.torchvision_version >>
+
 # -------------------------------------------------------------------------------------
 # Re-usable commands
 # -------------------------------------------------------------------------------------
+# install_nvidia_driver: &install_nvidia_driver
+#   - run:
+#       name: Install nvidia driver
+#       working_directory: ~/
+#       command: |
+#         wget -q 'https://s3.amazonaws.com/ossci-linux/nvidia_driver/NVIDIA-Linux-x86_64-430.40.run'
+#         sudo /bin/bash ./NVIDIA-Linux-x86_64-430.40.run -s --no-drm
+#         nvidia-smi
+
 install_python: &install_python
   - run:
       name: Install Python
       working_directory: ~/
       command: |
-        pyenv install 3.6.1
-        pyenv global 3.6.1
+        pyenv install -s 3.6.8
+        pyenv global 3.6.8
+        python --version
+        which python
+        pip install --upgrade pip setuptools==49.6.0
 
 setup_venv: &setup_venv
   - run:
@@ -60,50 +84,23 @@ setup_venv_win: &setup_venv_win
         which python
         which pip
 
-install_dep: &install_dep
+install_linux_dep: &install_linux_dep
   - run:
       name: Install Dependencies
       command: |
         pip install --progress-bar off -U 'git+https://github.com/facebookresearch/fvcore'
-        pip install --progress-bar off opencv-python pytest-xdist
-        pip install --progress-bar off torch torchvision
-
-install_dep_win: &install_dep_win
-  - run:
-      name: Install Dependencies on Windows
-      command: |
-        pip install certifi --ignore-installed
-        pip install numpy
-        pip install cython
-        pip install pycocotools>=2.0.2
-        pip install -U git+https://github.com/facebookresearch/fvcore
-        pip install opencv-python pytest-xdist
-        pip install torch==1.6.0+cpu torchvision==0.7.0+cpu -f https://download.pytorch.org/whl/torch_stable.html
+        pip install --progress-bar off ninja opencv-python-headless pytest-xdist tensorboard pycocotools
+        pip install --progress-bar off torch==$PYTORCH_VERSION torchvision==$TORCHVISION_VERSION -f https://download.pytorch.org/whl/cu101/torch_stable.html
+        python -c 'import torch; print("CUDA:", torch.cuda.is_available())'
+        gcc --version
 
 install_detectron2: &install_detectron2
   - run:
       name: Install Detectron2
       command: |
-        gcc --version
-        pip install -U --progress-bar off -e .[dev]
-        python -m detectron2.utils.collect_env
-
-install_detectron2_win: &install_detectron2_win
-  - run:
-      name: Install Detectron2 on Windows
-      command: |
-        pip install -U --progress-bar off -e .[dev]
+        pip install --progress-bar off -e .
         python -m detectron2.utils.collect_env
 
-install_nvidia_driver: &install_nvidia_driver
-  - run:
-      name: Install nvidia driver
-      working_directory: ~/
-      command: |
-        wget -q 'https://s3.amazonaws.com/ossci-linux/nvidia_driver/NVIDIA-Linux-x86_64-430.40.run'
-        sudo /bin/bash ./NVIDIA-Linux-x86_64-430.40.run -s --no-drm
-        nvidia-smi
-
 run_unittests: &run_unittests
   - run:
       name: Run Unit Tests
@@ -116,6 +113,7 @@ run_unittests: &run_unittests
 jobs:
   linux_cpu_tests:
     <<: *cpu
+    <<: *pytorch_parameters
 
     working_directory: ~/detectron2
 
@@ -126,92 +124,71 @@ jobs:
       # Cache the venv directory that contains dependencies
       - restore_cache:
           keys:
-            - cache-key-{{ .Branch }}-ID-20200730
+            - cache-{{ arch }}-<< parameters.pytorch_version >>-{{ .Branch }}-20200907
 
-      - <<: *install_dep
+      - <<: *install_linux_dep
 
       - save_cache:
           paths:
             - ~/venv
-          key: cache-key-{{ .Branch }}-ID-20200730
+          key: cache-{{ arch }}-<< parameters.pytorch_version >>-{{ .Branch }}-20200907
 
       - <<: *install_detectron2
       - <<: *run_unittests
 
   linux_gpu_tests:
     <<: *gpu
-
-    parameters:
-      pytorch_version:
-        type: string
-      torchvision_version:
-        type: string
-
-    environment:
-      PYTORCH_VERSION: << parameters.pytorch_version >>
-      TORCHVISION_VERSION: << parameters.torchvision_version >>
+    <<: *pytorch_parameters
 
     working_directory: ~/detectron2
 
     steps:
       - checkout
-      - <<: *install_nvidia_driver
 
-      - run:
-          name: Install nvidia-docker
-          working_directory: ~/
-          command: |
-            curl -s -L https://nvidia.github.io/nvidia-docker/gpgkey | sudo apt-key add -
-            distribution=$(. /etc/os-release;echo $ID$VERSION_ID)
-            curl -s -L https://nvidia.github.io/nvidia-docker/$distribution/nvidia-docker.list | \
-            sudo tee /etc/apt/sources.list.d/nvidia-docker.list
-            sudo apt-get update && sudo apt-get install -y nvidia-docker2
-            # reload the docker daemon configuration
-            sudo pkill -SIGHUP dockerd
+      # Cache the directory that contains python and dependencies
+      - restore_cache:
+          keys:
+            - cache-{{ arch }}-<< parameters.pytorch_version >>-{{ .Branch }}-20200907
 
-      - run:
-          name: Launch docker
-          working_directory: ~/detectron2/docker
-          command: |
-            nvidia-docker build -t detectron2:v0 -f Dockerfile-circleci --build-arg PYTORCH_VERSION=$PYTORCH_VERSION --build-arg TORCHVISION_VERSION=$TORCHVISION_VERSION .
-            nvidia-docker run -itd --name d2 detectron2:v0
-            docker exec -it d2 nvidia-smi
+      - <<: *install_python
+      - <<: *install_linux_dep
 
-      - run:
-          name: Build Detectron2
-          command: |
-            docker exec -it d2 pip install 'git+https://github.com/facebookresearch/fvcore'
-            docker cp ~/detectron2 d2:/detectron2
-            # This will build d2 for the target GPU arch only
-            # Install with --editable so that pytest can run tests in the source directory
-            docker exec -it d2 pip install --progress-bar off -e /detectron2
-            docker exec -it d2 python3 -m detectron2.utils.collect_env
-            docker exec -it d2 python3 -c 'import torch; assert(torch.cuda.is_available())'
+      - save_cache:
+          paths:
+            - /opt/circleci/.pyenv
+          key: cache-{{ arch }}-<< parameters.pytorch_version >>-{{ .Branch }}-20200907
 
-      - run:
-          name: Run Unit Tests
-          command: |
-            docker exec -e CIRCLECI=true -it d2 python3 -m pytest -n 2 -v /detectron2/tests
+      - <<: *install_detectron2
+      - <<: *run_unittests
 
   windows_cpu_tests:
     <<: *windows_cpu
+    <<: *pytorch_parameters
     steps:
       - checkout
       - <<: *setup_venv_win
 
       # Cache the env directory that contains dependencies
       - restore_cache:
           keys:
-            - cache-key-{{ .Branch }}-ID-20200735
+            - cache-{{ arch }}-<< parameters.pytorch_version >>-{{ .Branch }}-20200907
 
-      - <<: *install_dep_win
+      - run:
+          name: Install Dependencies
+          command: |
+            pip install certifi --ignore-installed  # required on windows to workaround some cert issue
+            pip install numpy cython  # required on windows before pycocotools
+            pip install opencv-python-headless pytest-xdist pycocotools tensorboard
+            pip install -U git+https://github.com/facebookresearch/fvcore
+            pip install torch==1.6.0+cpu torchvision==0.7.0+cpu -f https://download.pytorch.org/whl/torch_stable.html
 
       - save_cache:
           paths:
             - env
-          key: cache-key-{{ .Branch }}-ID-20200735
+          key: cache-{{ arch }}-<< parameters.pytorch_version >>-{{ .Branch }}-20200907
 
-      - <<: *install_detectron2_win
+      - <<: *install_detectron2
+      # TODO: unittest fails for now
 
 workflows:
   version: 2

.github/workflows/workflow.yml

@@ -58,9 +58,8 @@ jobs:
 
       - name: Install dependencies
         run: |
-          brew install ninja
           python -m pip install -U pip setuptools==49
-          python -m pip install opencv-python-headless onnx pytest-xdist
+          python -m pip install ninja opencv-python-headless onnx pytest-xdist
           python -m pip install torch==${{matrix.torch}} torchvision==${{matrix.torchvision}} -f https://download.pytorch.org/whl/torch_stable.html
           python -m pip install -U 'git+https://github.com/facebookresearch/fvcore'
 

INSTALL.md

@@ -146,18 +146,18 @@ Two possibilities:
   you need to either install a different build of PyTorch (or build by yourself)
   to match your local CUDA installation, or install a different version of CUDA to match PyTorch.
 
-* PyTorch/torchvision/Detectron2 is not built for the correct GPU architecture (aka. compute compatibility).
+* PyTorch/torchvision/Detectron2 is not built for the correct GPU architecture (aka. compute capability).
 
-  The compute compatibility included by PyTorch/detectron2/torchvision is available in the "architecture flags" in
+  The architecture included by PyTorch/detectron2/torchvision is available in the "architecture flags" in
   `python -m detectron2.utils.collect_env`. It must include
-  the compute compatibility of your GPU, which can be found at [developer.nvidia.com/cuda-gpus](https://developer.nvidia.com/cuda-gpus).
+  the architecture of your GPU, which can be found at [developer.nvidia.com/cuda-gpus](https://developer.nvidia.com/cuda-gpus).
 
   If you're using pre-built PyTorch/detectron2/torchvision, they have included support for most popular GPUs already.
   If not supported, you need to build them from source.
 
   When building detectron2/torchvision from source, they detect the GPU device and build for only the device.
   This means the compiled code may not work on a different GPU device.
-  To recompile them for the correct compatiblity, remove all installed/compiled files,
+  To recompile them for the correct architecture, remove all installed/compiled files,
   and rebuild them with the `TORCH_CUDA_ARCH_LIST` environment variable set properly.
   For example, `export TORCH_CUDA_ARCH_LIST=6.0,7.0` makes it compile for both P100s and V100s.
 </details>

detectron2/evaluation/coco_evaluation.py

@@ -118,7 +118,6 @@ def process(self, inputs, outputs):
         for input, output in zip(inputs, outputs):
             prediction = {"image_id": input["image_id"]}
 
-            # TODO this is ugly
             if "instances" in output:
                 instances = output["instances"].to(self._cpu_device)
                 prediction["instances"] = instances_to_coco_json(instances, input["image_id"])

detectron2/utils/collect_env.py

@@ -185,3 +185,11 @@ def collect_env_info():
         from detectron2.utils.collect_env import collect_env_info
 
         print(collect_env_info())
+    if torch.cuda.is_available():
+        for k in range(torch.cuda.device_count()):
+            device = f"cuda:{k}"
+            try:
+                x = torch.tensor([1, 2.0], dtype=torch.float32)
+                x = x.to(device)
+            except Exception:
+                print(f"Unable to copy tensor to device={device}")