Remove lots of code

Author: NicolasHug

.github/scripts/setup-env.sh

@@ -23,15 +23,13 @@ case $(uname) in
 esac
 
 echo '::group::Create build environment'
-# See https://github.com/pytorch/vision/issues/7296 for ffmpeg
 conda create \
   --name ci \
   --quiet --yes \
   python="${PYTHON_VERSION}" pip \
   ninja cmake \
   libpng \
   libwebp \
-  'ffmpeg<4.3'
 conda activate ci
 conda install --quiet --yes libjpeg-turbo -c pytorch
 pip install --progress-bar=off --upgrade setuptools==72.1.0

.github/scripts/unittest.sh

@@ -14,5 +14,4 @@ echo '::endgroup::'
 
 python test/smoke_test.py
 
-# We explicitly ignore the video tests until we resolve https://github.com/pytorch/vision/issues/8162
-pytest --ignore-glob="*test_video*" --ignore-glob="*test_onnx*" --junit-xml="${RUNNER_TEST_RESULTS_DIR}/test-results.xml" -v --durations=25 -k "not TestFxFeatureExtraction"
+pytest --ignore-glob="*test_onnx*" --junit-xml="${RUNNER_TEST_RESULTS_DIR}/test-results.xml" -v --durations=25 -k "not TestFxFeatureExtraction"

.github/workflows/docs.yml

@@ -34,12 +34,6 @@ jobs:
         CONDA_PATH=$(which conda)
         eval "$(${CONDA_PATH} shell.bash hook)"
         conda activate ci
-        # FIXME: not sure why we need this. `ldd torchvision/video_reader.so` shows that it
-        #  already links against the one pulled from conda. However, at runtime it pulls from
-        #  /lib64
-        # Should we maybe always do this in `./.github/scripts/setup-env.sh` so that we don't
-        # have to pay attention in all other workflows?
-        export LD_LIBRARY_PATH="${CONDA_PREFIX}/lib:${LD_LIBRARY_PATH}"
 
         cd docs
 

docs/source/io.rst

@@ -1,10 +1,10 @@
-Decoding / Encoding images and videos
-=====================================
+Decoding / Encoding images
+==========================
 
 .. currentmodule:: torchvision.io
 
 The :mod:`torchvision.io` module provides utilities for decoding and encoding
-images and videos.
+images.
 
 Image Decoding
 --------------
@@ -92,7 +92,7 @@ Video - DEPREACTED
 .. warning::
 
     DEPRECATED: All the video decoding and encoding capabilities of torchvision
-    are deprecated from version 0.22 and will be removed in version 0.24.  We
+    are deprecated from version 0.22 and will be removed in version 0.25.  We
     recommend that you migrate to
     `TorchCodec <https://github.com/pytorch/torchcodec>`__, where we'll
     consolidate the future decoding/encoding capabilities of PyTorch
@@ -101,19 +101,4 @@ Video - DEPREACTED
     :toctree: generated/
     :template: function.rst
 
-    read_video
-    read_video_timestamps
     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.
-
-.. autosummary::
-    :toctree: generated/
-    :template: class.rst
-
-    VideoReader

mypy.ini

@@ -41,10 +41,6 @@ ignore_errors = True
 
 ignore_errors = True
 
-[mypy-torchvision.io.video_reader]
-
-ignore_errors = True
-
 [mypy-torchvision.models.*]
 
 ignore_errors=True

packaging/pre_build_script.sh

@@ -17,7 +17,6 @@ if [[ "$(uname)" == Darwin || "$OSTYPE" == "msys" ]]; then
   # Installing webp also installs a non-turbo jpeg, so we uninstall jpeg stuff
   # before re-installing them
   conda uninstall libjpeg-turbo libjpeg -y
-  conda install -y ffmpeg=4.2 -c pytorch
   conda install -y libjpeg-turbo -c pytorch
 
   # Copy binaries to be included in the wheel distribution
@@ -30,7 +29,7 @@ else
 
   if [[ "$ARCH" == "aarch64" ]]; then
     conda install libpng -y
-    conda install -y ffmpeg=4.2 libjpeg-turbo -c pytorch-nightly
+    conda install -y libjpeg-turbo -c pytorch-nightly
   fi
 
   conda install libwebp -y

packaging/wheel/relocate.py

@@ -316,8 +316,7 @@ def patch_linux():
     output_dir = osp.join(PACKAGE_ROOT, "dist", ".wheel-process")
 
     image_binary = "image.so"
-    video_binary = "video_reader.so"
-    torchvision_binaries = [image_binary, video_binary]
+    torchvision_binaries = [image_binary, ]
     for wheel in wheels:
         if osp.exists(output_dir):
             shutil.rmtree(output_dir)
@@ -352,8 +351,7 @@ def patch_win():
     output_dir = osp.join(PACKAGE_ROOT, "dist", ".wheel-process")
 
     image_binary = "image.pyd"
-    video_binary = "video_reader.pyd"
-    torchvision_binaries = [image_binary, video_binary]
+    torchvision_binaries = [image_binary, ]
     for wheel in wheels:
         if osp.exists(output_dir):
             shutil.rmtree(output_dir)

setup.py

@@ -22,14 +22,6 @@
 USE_WEBP = os.getenv("TORCHVISION_USE_WEBP", "1") == "1"
 USE_NVJPEG = os.getenv("TORCHVISION_USE_NVJPEG", "1") == "1"
 NVCC_FLAGS = os.getenv("NVCC_FLAGS", None)
-# Note: the GPU video decoding stuff used to be called "video codec", which
-# isn't an accurate or descriptive name considering there are at least 2 other
-# video decoding backends in torchvision. I'm renaming this to "gpu video
-# decoder" where possible, keeping user facing names (like the env var below) to
-# the old scheme for BC.
-USE_GPU_VIDEO_DECODER = os.getenv("TORCHVISION_USE_VIDEO_CODEC", "1") == "1"
-# Same here: "use ffmpeg" was used to denote "use cpu video decoder".
-USE_CPU_VIDEO_DECODER = os.getenv("TORCHVISION_USE_FFMPEG", "1") == "1"
 
 TORCHVISION_INCLUDE = os.environ.get("TORCHVISION_INCLUDE", "")
 TORCHVISION_LIBRARY = os.environ.get("TORCHVISION_LIBRARY", "")
@@ -52,8 +44,6 @@
 print(f"{USE_WEBP = }")
 print(f"{USE_NVJPEG = }")
 print(f"{NVCC_FLAGS = }")
-print(f"{USE_CPU_VIDEO_DECODER = }")
-print(f"{USE_GPU_VIDEO_DECODER = }")
 print(f"{TORCHVISION_INCLUDE = }")
 print(f"{TORCHVISION_LIBRARY = }")
 print(f"{IS_ROCM = }")
@@ -371,158 +361,6 @@ def make_image_extension():
     )
 
 
-def make_video_decoders_extensions():
-    print("Building video decoder extensions")
-
-    build_without_extensions_msg = "Building without video decoders extensions."
-    if sys.platform != "linux" or (sys.version_info.major == 3 and sys.version_info.minor == 9):
-        # FIXME: Building torchvision with ffmpeg on MacOS or with Python 3.9
-        # FIXME: causes crash. See the following GitHub issues for more details.
-        # FIXME: https://github.com/pytorch/pytorch/issues/65000
-        # FIXME: https://github.com/pytorch/vision/issues/3367
-        print("Can only build video decoder extensions on linux and Python != 3.9")
-        return []
-
-    ffmpeg_exe = shutil.which("ffmpeg")
-    if ffmpeg_exe is None:
-        print(f"{build_without_extensions_msg} Couldn't find ffmpeg binary.")
-        return []
-
-    def find_ffmpeg_libraries():
-        ffmpeg_libraries = {"libavcodec", "libavformat", "libavutil", "libswresample", "libswscale"}
-
-        ffmpeg_bin = os.path.dirname(ffmpeg_exe)
-        ffmpeg_root = os.path.dirname(ffmpeg_bin)
-        ffmpeg_include_dir = os.path.join(ffmpeg_root, "include")
-        ffmpeg_library_dir = os.path.join(ffmpeg_root, "lib")
-
-        gcc = os.environ.get("CC", shutil.which("gcc"))
-        platform_tag = subprocess.run([gcc, "-print-multiarch"], stdout=subprocess.PIPE)
-        platform_tag = platform_tag.stdout.strip().decode("utf-8")
-
-        if platform_tag:
-            # Most probably a Debian-based distribution
-            ffmpeg_include_dir = [ffmpeg_include_dir, os.path.join(ffmpeg_include_dir, platform_tag)]
-            ffmpeg_library_dir = [ffmpeg_library_dir, os.path.join(ffmpeg_library_dir, platform_tag)]
-        else:
-            ffmpeg_include_dir = [ffmpeg_include_dir]
-            ffmpeg_library_dir = [ffmpeg_library_dir]
-
-        for library in ffmpeg_libraries:
-            library_found = False
-            for search_path in ffmpeg_include_dir + TORCHVISION_INCLUDE:
-                full_path = os.path.join(search_path, library, "*.h")
-                library_found |= len(glob.glob(full_path)) > 0
-
-            if not library_found:
-                print(f"{build_without_extensions_msg}")
-                print(f"{library} header files were not found.")
-                return None, None
-
-        return ffmpeg_include_dir, ffmpeg_library_dir
-
-    ffmpeg_include_dir, ffmpeg_library_dir = find_ffmpeg_libraries()
-    if ffmpeg_include_dir is None or ffmpeg_library_dir is None:
-        return []
-
-    print("Found ffmpeg:")
-    print(f"  ffmpeg include path: {ffmpeg_include_dir}")
-    print(f"  ffmpeg library_dir: {ffmpeg_library_dir}")
-
-    extensions = []
-    if USE_CPU_VIDEO_DECODER:
-        print("Building with CPU video decoder support")
-
-        # TorchVision base decoder + video reader
-        video_reader_src_dir = os.path.join(ROOT_DIR, "torchvision", "csrc", "io", "video_reader")
-        video_reader_src = glob.glob(os.path.join(video_reader_src_dir, "*.cpp"))
-        base_decoder_src_dir = os.path.join(ROOT_DIR, "torchvision", "csrc", "io", "decoder")
-        base_decoder_src = glob.glob(os.path.join(base_decoder_src_dir, "*.cpp"))
-        # Torchvision video API
-        videoapi_src_dir = os.path.join(ROOT_DIR, "torchvision", "csrc", "io", "video")
-        videoapi_src = glob.glob(os.path.join(videoapi_src_dir, "*.cpp"))
-        # exclude tests
-        base_decoder_src = [x for x in base_decoder_src if "_test.cpp" not in x]
-
-        combined_src = video_reader_src + base_decoder_src + videoapi_src
-
-        extensions.append(
-            CppExtension(
-                # This is an awful name. It should be "cpu_video_decoder". Keeping for BC.
-                "torchvision.video_reader",
-                combined_src,
-                include_dirs=[
-                    base_decoder_src_dir,
-                    video_reader_src_dir,
-                    videoapi_src_dir,
-                    str(CSRS_DIR),
-                    *ffmpeg_include_dir,
-                    *TORCHVISION_INCLUDE,
-                ],
-                library_dirs=ffmpeg_library_dir + TORCHVISION_LIBRARY,
-                libraries=[
-                    "avcodec",
-                    "avformat",
-                    "avutil",
-                    "swresample",
-                    "swscale",
-                ],
-                extra_compile_args=["-std=c++17"] if os.name != "nt" else ["/std:c++17", "/MP"],
-                extra_link_args=["-std=c++17" if os.name != "nt" else "/std:c++17"],
-            )
-        )
-
-    if USE_GPU_VIDEO_DECODER:
-        # Locating GPU video decoder headers and libraries
-        # CUDA_HOME should be set to the cuda root directory.
-        # TORCHVISION_INCLUDE and TORCHVISION_LIBRARY should include the locations
-        # to the headers and libraries below
-        if not (
-            BUILD_CUDA_SOURCES
-            and CUDA_HOME is not None
-            and any([os.path.exists(os.path.join(folder, "cuviddec.h")) for folder in TORCHVISION_INCLUDE])
-            and any([os.path.exists(os.path.join(folder, "nvcuvid.h")) for folder in TORCHVISION_INCLUDE])
-            and any([os.path.exists(os.path.join(folder, "libnvcuvid.so")) for folder in TORCHVISION_LIBRARY])
-            and any([os.path.exists(os.path.join(folder, "libavcodec", "bsf.h")) for folder in ffmpeg_include_dir])
-        ):
-            print("Could not find necessary dependencies. Refer the setup.py to check which ones are needed.")
-            print("Building without GPU video decoder support")
-            return extensions
-        print("Building torchvision with GPU video decoder support")
-
-        gpu_decoder_path = os.path.join(CSRS_DIR, "io", "decoder", "gpu")
-        gpu_decoder_src = glob.glob(os.path.join(gpu_decoder_path, "*.cpp"))
-        cuda_libs = os.path.join(CUDA_HOME, "lib64")
-        cuda_inc = os.path.join(CUDA_HOME, "include")
-
-        _, extra_compile_args = get_macros_and_flags()
-        extensions.append(
-            CUDAExtension(
-                "torchvision.gpu_decoder",
-                gpu_decoder_src,
-                include_dirs=[CSRS_DIR] + TORCHVISION_INCLUDE + [gpu_decoder_path] + [cuda_inc] + ffmpeg_include_dir,
-                library_dirs=ffmpeg_library_dir + TORCHVISION_LIBRARY + [cuda_libs],
-                libraries=[
-                    "avcodec",
-                    "avformat",
-                    "avutil",
-                    "swresample",
-                    "swscale",
-                    "nvcuvid",
-                    "cuda",
-                    "cudart",
-                    "z",
-                    "pthread",
-                    "dl",
-                    "nppicc",
-                ],
-                extra_compile_args=extra_compile_args,
-            )
-        )
-
-    return extensions
-
-
 class clean(distutils.command.clean.clean):
     def run(self):
         with open(".gitignore") as f:
@@ -550,7 +388,6 @@ def run(self):
     extensions = [
         make_C_extension(),
         make_image_extension(),
-        *make_video_decoders_extensions(),
     ]
 
     setup(

torchvision/__init__.py

@@ -28,8 +28,6 @@
 
 _image_backend = "PIL"
 
-_video_backend = "pyav"
-
 
 def set_image_backend(backend):
     """
@@ -53,48 +51,6 @@ def get_image_backend():
     return _image_backend
 
 
-def set_video_backend(backend):
-    """
-    Specifies the package used to decode videos.
-
-    Args:
-        backend (string): Name of the video backend. one of {'pyav', 'video_reader'}.
-            The :mod:`pyav` package uses the 3rd party PyAv library. It is a Pythonic
-            binding for the FFmpeg libraries.
-            The :mod:`video_reader` package includes a native C++ implementation on
-            top of FFMPEG libraries, and a python API of TorchScript custom operator.
-            It generally decodes faster than :mod:`pyav`, but is perhaps less robust.
-
-    .. note::
-        Building with FFMPEG is disabled by default in the latest `main`. If you want to use the 'video_reader'
-        backend, please compile torchvision from source.
-    """
-    global _video_backend
-    if backend not in ["pyav", "video_reader", "cuda"]:
-        raise ValueError("Invalid video backend '%s'. Options are 'pyav', 'video_reader' and 'cuda'" % backend)
-    if backend == "video_reader" and not io._HAS_CPU_VIDEO_DECODER:
-        # TODO: better messages
-        message = "video_reader video backend is not available. Please compile torchvision from source and try again"
-        raise RuntimeError(message)
-    elif backend == "cuda" and not io._HAS_GPU_VIDEO_DECODER:
-        # TODO: better messages
-        message = "cuda video backend is not available."
-        raise RuntimeError(message)
-    else:
-        _video_backend = backend
-
-
-def get_video_backend():
-    """
-    Returns the currently active video backend used to decode videos.
-
-    Returns:
-        str: Name of the video backend. one of {'pyav', 'video_reader'}.
-    """
-
-    return _video_backend
-
-
 def _is_tracing():
     return torch._C._get_tracing_state()