Merge branch 'master' into master

Author: alihassanijr

.circleci/config.yml

@@ -1145,6 +1145,10 @@ workflows:
           cu_version: cpu
           name: cmake_windows_cpu
           python_version: '3.8'
+      - cmake_windows_gpu:
+          cu_version: cu101
+          name: cmake_windows_gpu
+          python_version: '3.8'
       - cmake_macos_cpu:
           cu_version: cpu
           name: cmake_macos_cpu

.circleci/regenerate.py

@@ -191,16 +191,16 @@ def cmake_workflows(indentation=6):
     jobs = []
     python_version = '3.8'
     for os_type in ['linux', 'windows', 'macos']:
-        # Right now CMake builds are failling on Windows (GPU)
-        device_types = ['cpu', 'gpu'] if os_type == 'linux' else ['cpu']
+        # Skip OSX CUDA
+        device_types = ['cpu', 'gpu'] if os_type != 'macos' else ['cpu']
         for device in device_types:
             job = {
                 'name': f'cmake_{os_type}_{device}',
                 'python_version': python_version
             }
 
             job['cu_version'] = 'cu101' if device == 'gpu' else 'cpu'
-            if device == 'gpu':
+            if device == 'gpu' and os_type == 'linux':
                 job['wheel_docker_image'] = 'pytorch/manylinux-cuda101'
             jobs.append({f'cmake_{os_type}_{device}': job})
     return indent(indentation, jobs)

.circleci/unittest/linux/scripts/environment.yml

@@ -1,4 +1,5 @@
 channels:
+  - pytorch
   - defaults
 dependencies:
   - numpy
@@ -8,6 +9,7 @@ dependencies:
   - pip
   - libpng
   - jpeg
+  - ffmpeg=4.2
   - ca-certificates
   - pip:
     - future

.circleci/unittest/windows/scripts/environment.yml

@@ -1,4 +1,5 @@
 channels:
+  - pytorch
   - defaults
 dependencies:
   - numpy

.gitattributes

@@ -1 +1,8 @@
 *.pkl binary
+# Jupyter notebook
+
+# For text count
+*.ipynb text
+
+# To ignore it use below
+# *.ipynb linguist-documentation

.github/workflows/tests-schedule.yml

@@ -35,7 +35,7 @@ jobs:
         run: pip install pytest
 
       - name: Run tests
-        run: pytest --durations=20 -ra test/test_datasets_download.py
+        run: pytest -ra -v test/test_datasets_download.py
 
       - uses: JasonEtco/[email protected]
         name: Create issue if download tests failed

.gitignore

@@ -23,3 +23,4 @@ gen.yml
 .mypy_cache
 .vscode/
 *.orig
+*-checkpoint.ipynb

.travis.yml

@@ -38,7 +38,6 @@ before_install:
     fi
   - conda install av -c conda-forge
 
-
 install:
   # Using pip instead of setup.py ensures we install a non-compressed version of the package
   # (as opposed to an egg), which is necessary to collect coverage.
@@ -55,7 +54,7 @@ install:
     cd -
 
 script:
-  - pytest --cov-config .coveragerc --cov torchvision --cov $TV_INSTALL_PATH -k 'not TestVideoReader and not TestVideoTransforms and not TestIO' test --ignore=test/test_datasets_download.py
+  - pytest --cov-config .coveragerc --cov torchvision --cov $TV_INSTALL_PATH -k 'not TestVideo and not TestVideoReader and not TestVideoTransforms and not TestIO' test --ignore=test/test_datasets_download.py
   - pytest test/test_hub.py
 
 after_success:

CMakeLists.txt

@@ -9,6 +9,7 @@ if(WITH_CUDA)
   enable_language(CUDA)
   add_definitions(-D__CUDA_NO_HALF_OPERATORS__)
   add_definitions(-DWITH_CUDA)
+  set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} --expt-relaxed-constexpr")
 endif()
 
 find_package(Python3 COMPONENTS Development)
@@ -17,6 +18,15 @@ find_package(Torch REQUIRED)
 find_package(PNG REQUIRED)
 find_package(JPEG REQUIRED)
 
+function(CUDA_CONVERT_FLAGS EXISTING_TARGET)
+    get_property(old_flags TARGET ${EXISTING_TARGET} PROPERTY INTERFACE_COMPILE_OPTIONS)
+    if(NOT "${old_flags}" STREQUAL "")
+        string(REPLACE ";" "," CUDA_flags "${old_flags}")
+        set_property(TARGET ${EXISTING_TARGET} PROPERTY INTERFACE_COMPILE_OPTIONS
+            "$<$<BUILD_INTERFACE:$<COMPILE_LANGUAGE:CXX>>:${old_flags}>$<$<BUILD_INTERFACE:$<COMPILE_LANGUAGE:CUDA>>:-Xcompiler=${CUDA_flags}>"
+            )
+    endif()
+endfunction()
 
 file(GLOB HEADERS torchvision/csrc/*.h)
 # Image extension
@@ -29,6 +39,26 @@ endif()
 file(GLOB MODELS_HEADERS torchvision/csrc/models/*.h)
 file(GLOB MODELS_SOURCES torchvision/csrc/models/*.h torchvision/csrc/models/*.cpp)
 
+if(MSVC)
+  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /wd4819")
+  if(WITH_CUDA)
+    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Xcompiler=/wd4819")
+    foreach(diag cc_clobber_ignored integer_sign_change useless_using_declaration
+      set_but_not_used field_without_dll_interface
+      base_class_has_different_dll_interface
+      dll_interface_conflict_none_assumed
+      dll_interface_conflict_dllexport_assumed
+      implicit_return_from_non_void_function
+      unsigned_compare_with_zero
+      declared_but_not_referenced
+      bad_friend_decl)
+      string(APPEND CMAKE_CUDA_FLAGS " -Xcudafe --diag_suppress=${diag}")
+    endforeach()
+    CUDA_CONVERT_FLAGS(torch_cpu)
+    CUDA_CONVERT_FLAGS(torch_cuda)
+  endif()
+endif()
+
 add_library(${PROJECT_NAME} SHARED ${MODELS_SOURCES} ${OPERATOR_SOURCES} ${IMAGE_SOURCES})
 target_link_libraries(${PROJECT_NAME} PRIVATE ${TORCH_LIBRARIES} ${PNG_LIBRARY} ${JPEG_LIBRARIES} Python3::Python)
 set_target_properties(${PROJECT_NAME} PROPERTIES EXPORT_NAME TorchVision)

cmake/TorchVisionConfig.cmake.in

@@ -32,7 +32,7 @@ if(NOT TARGET Python3::Python)
 find_package(Python3 COMPONENTS Development)
 endif()
 
-set_target_properties(TorchVision::TorchVision PROPERTIES INTERFACE_INCLUDE_DIRECTORIES "${PACKAGE_PREFIX_DIR}/@CMAKE_INSTALL_INCLUDEDIR@" INTERFACE_LINK_LIBRARIES "torch;Python3::Python" )
+set_target_properties(TorchVision::TorchVision PROPERTIES INTERFACE_INCLUDE_DIRECTORIES "${${PN}_INCLUDE_DIR}" INTERFACE_LINK_LIBRARIES "torch;Python3::Python" )
 
 
 if(@WITH_CUDA@)

docs/source/io.rst

@@ -4,7 +4,8 @@ torchvision.io
 .. currentmodule:: torchvision.io
 
 The :mod:`torchvision.io` package provides functions for performing IO
-operations. They are currently specific to reading and writing video.
+operations. They are currently specific to reading and writing video and
+images.
 
 Video
 -----
@@ -14,3 +15,58 @@ Video
 .. autofunction:: read_video_timestamps
 
 .. autofunction:: write_video
+
+
+Fine-grained video API
+-------------------
+
+In addition to the :mod:`read_video` function, we provide a high-performance 
+lower-level API for more fine-grained control compared to the :mod:`read_video` function.
+It does all this whilst fully supporting torchscript.
+
+.. autoclass:: VideoReader
+    :members: __next__, get_metadata, set_current_stream, seek
+
+
+Example of inspecting a video:
+
+.. code:: python
+
+    import torchvision
+    video_path = "path to a test video"
+    # Constructor allocates memory and a threaded decoder
+    # instance per video. At the momet it takes two arguments:
+    # path to the video file, and a wanted stream.
+    reader = torchvision.io.VideoReader(video_path, "video")
+
+    # The information about the video can be retrieved using the 
+    # `get_metadata()` method. It returns a dictionary for every stream, with
+    # duration and other relevant metadata (often frame rate)
+    reader_md = reader.get_metadata()
+
+    # metadata is structured as a dict of dicts with following structure
+    # {"stream_type": {"attribute": [attribute per stream]}}
+    #
+    # following would print out the list of frame rates for every present video stream
+    print(reader_md["video"]["fps"])
+
+    # we explicitly select the stream we would like to operate on. In
+    # the constructor we select a default video stream, but
+    # in practice, we can set whichever stream we would like 
+    video.set_current_stream("video:0")
+
+
+Image
+-----
+
+.. autofunction:: read_image
+
+.. autofunction:: decode_image
+
+.. autofunction:: encode_jpeg
+
+.. autofunction:: write_jpeg
+
+.. autofunction:: encode_png
+
+.. autofunction:: write_png

docs/source/models.rst

@@ -350,6 +350,7 @@ the instances set of COCO train2017 and evaluated on COCO val2017.
 Network                           box AP   mask AP   keypoint AP
 ================================  =======  ========  ===========
 Faster R-CNN ResNet-50 FPN        37.0     -         -
+RetinaNet ResNet-50 FPN           36.4     -         -
 Mask R-CNN ResNet-50 FPN          37.9     34.6      -
 ================================  =======  ========  ===========
 
@@ -405,6 +406,7 @@ precision-recall.
 Network                         train time (s / it)  test time (s / it)  memory (GB)
 ==============================  ===================  ==================  ===========
 Faster R-CNN ResNet-50 FPN      0.2288               0.0590              5.2
+RetinaNet ResNet-50 FPN         0.2514               0.0939              4.1
 Mask R-CNN ResNet-50 FPN        0.2728               0.0903              5.4
 Keypoint R-CNN ResNet-50 FPN    0.3789               0.1242              6.8
 ==============================  ===================  ==================  ===========
@@ -416,6 +418,12 @@ Faster R-CNN
 .. autofunction:: torchvision.models.detection.fasterrcnn_resnet50_fpn
 
 
+RetinaNet
+------------
+
+.. autofunction:: torchvision.models.detection.retinanet_resnet50_fpn
+
+
 Mask R-CNN
 ----------
 

docs/source/transforms.rst

@@ -14,9 +14,27 @@ All transformations accept PIL Image, Tensor Image or batch of Tensor Images as
 Tensor Images is a tensor of ``(B, C, H, W)`` shape, where ``B`` is a number of images in the batch. Deterministic or
 random transformations applied on the batch of Tensor Images identically transform all the images of the batch.
 
+.. warning::
+
+    Since v0.8.0 all random transformations are using torch default random generator to sample random parameters.
+    It is a backward compatibility breaking change and user should set the random state as following:
+
+    .. code:: python
+
+        # Previous versions
+        # import random
+        # random.seed(12)
+
+        # Now
+        import torch
+        torch.manual_seed(17)
+
+    Please, keep in mind that the same seed for torch random generator and Python random generator will not
+    produce the same results.
+
 
 Scriptable transforms
-^^^^^^^^^^^^^^^^^^^^^
+---------------------
 
 In order to script the transformations, please use ``torch.nn.Sequential`` instead of :class:`Compose`.
 
@@ -34,6 +52,9 @@ Make sure to use only scriptable transformations, i.e. that work with ``torch.Te
 For any custom transformations to be used with ``torch.jit.script``, they should be derived from ``torch.nn.Module``.
 
 
+Compositions of transforms
+--------------------------
+
 .. autoclass:: Compose
 
 Transforms on PIL Image
@@ -79,6 +100,8 @@ Transforms on PIL Image
 
 .. autoclass:: TenCrop
 
+.. autoclass:: GaussianBlur
+
 Transforms on torch.\*Tensor
 ----------------------------
 

examples/python/README.md

@@ -0,0 +1,18 @@
+# Python examples
+
+- [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/pytorch/vision/blob/master/examples/python/tensor_transforms.ipynb)
+[Examples of Tensor Images transformations](https://github.com/pytorch/vision/blob/master/examples/python/tensor_transforms.ipynb)
+- [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/pytorch/vision/blob/master/examples/python/video_api.ipynb)
+[Example of VideoAPI](https://github.com/pytorch/vision/blob/master/examples/python/video_api.ipynb)
+
+
+Prior to v0.8.0, transforms in torchvision have traditionally been PIL-centric and presented multiple limitations due to 
+that. Now, since v0.8.0, transforms implementations are Tensor and PIL compatible and we can achieve the following new 
+features:
+- transform multi-band torch tensor images (with more than 3-4 channels) 
+- torchscript transforms together with your model for deployment
+- support for GPU acceleration
+- batched transformation such as for videos
+- read and decode data directly as torch tensor with torchscript support (for PNG and JPEG image formats)
+
+Furthermore, previously we used to provide a very high-level API for video decoding which left little control to the user. We're now expanding that API (and replacing it in the future) with a lower-level API that allows the user a frame-based access to a video.