Merge branch 'main' into feature/different-label-background

Author: AndreiMoraru123

.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'
+  libwebp
 conda activate ci
 conda install --quiet --yes libjpeg-turbo -c pytorch
 pip install --progress-bar=off --upgrade setuptools==72.1.0

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

setup.py

@@ -28,9 +28,9 @@
 # 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"
+USE_GPU_VIDEO_DECODER = os.getenv("TORCHVISION_USE_VIDEO_CODEC", "0") == "1"
 # Same here: "use ffmpeg" was used to denote "use cpu video decoder".
-USE_CPU_VIDEO_DECODER = os.getenv("TORCHVISION_USE_FFMPEG", "1") == "1"
+USE_CPU_VIDEO_DECODER = os.getenv("TORCHVISION_USE_FFMPEG", "0") == "1"
 
 TORCHVISION_INCLUDE = os.environ.get("TORCHVISION_INCLUDE", "")
 TORCHVISION_LIBRARY = os.environ.get("TORCHVISION_LIBRARY", "")

torchvision/csrc/ops/cuda/nms_kernel.cu

@@ -34,8 +34,8 @@ __global__ void nms_kernel_impl(
     double iou_threshold,
     const T* dev_boxes,
     unsigned long long* dev_mask) {
-  const int row_start = blockIdx.y;
-  const int col_start = blockIdx.x;
+  const auto row_start = blockIdx.y;
+  const auto col_start = blockIdx.x;
 
   if (row_start > col_start)
     return;
@@ -59,7 +59,7 @@ __global__ void nms_kernel_impl(
   __syncthreads();
 
   if (threadIdx.x < row_size) {
-    const int cur_box_idx = threadsPerBlock * row_start + threadIdx.x;
+    const auto cur_box_idx = threadsPerBlock * row_start + threadIdx.x;
     const T* cur_box = dev_boxes + cur_box_idx * 4;
     int i = 0;
     unsigned long long t = 0;
@@ -84,7 +84,7 @@ __global__ static void gather_keep_from_mask(
   // Taken and adapted from mmcv
   // https://github.com/open-mmlab/mmcv/blob/03ce9208d18c0a63d7ffa087ea1c2f5661f2441a/mmcv/ops/csrc/common/cuda/nms_cuda_kernel.cuh#L76
   const int col_blocks = ceil_div(n_boxes, threadsPerBlock);
-  const int thread_id = threadIdx.x;
+  const auto thread_id = threadIdx.x;
 
   // Mark the bboxes which have been removed.
   extern __shared__ unsigned long long removed[];