paddlecodec 0.1.0

A video decoder for PyTorch

Installation | Simple Example | Detailed Example | Documentation | Contributing | License

PaddleCodec

[!NOTE]

This repo is a fork of the original torchcodec project, with modifications to enhance compatibility and integration with PaddlePaddle.

Installation

pip install paddlecodec

Usage

import paddle
paddle.enable_compat(scope={"torchcodec"})  # Enable torch proxy before importing torchcodec
import torchcodec
# use torchcodec

The original README.md content is as follows:

---

TorchCodec is a Python library for decoding video and audio data into PyTorch tensors, on CPU and CUDA GPU. It also supports audio encoding, and video encoding will come soon! It aims to be fast, easy to use, and well integrated into the PyTorch ecosystem. If you want to use PyTorch to train ML models on videos and audio, TorchCodec is how you turn these into data.

We achieve these capabilities through:

• Pythonic APIs that mirror Python and PyTorch conventions.
• Relying on FFmpeg to do the decoding and encoding. TorchCodec uses the version of FFmpeg you already have installed. FFmpeg is a mature library with broad coverage available on most systems. It is, however, not easy to use. TorchCodec abstracts FFmpeg's complexity to ensure it is used correctly and efficiently.
• Returning data as PyTorch tensors, ready to be fed into PyTorch transforms or used directly to train models.

Using TorchCodec

Here's a condensed summary of what you can do with TorchCodec. For more detailed examples, check out our documentation!

Decoding

from torchcodec.decoders import VideoDecoder

device = "cpu"  # or e.g. "cuda" !
decoder = VideoDecoder("path/to/video.mp4", device=device)

decoder.metadata
# VideoStreamMetadata:
#   num_frames: 250
#   duration_seconds: 10.0
#   bit_rate: 31315.0
#   codec: h264
#   average_fps: 25.0
#   ... (truncated output)

# Simple Indexing API
decoder[0]  # uint8 tensor of shape [C, H, W]
decoder[0 : -1 : 20]  # uint8 stacked tensor of shape [N, C, H, W]

# Indexing, with PTS and duration info:
decoder.get_frames_at(indices=[2, 100])
# FrameBatch:
#   data (shape): torch.Size([2, 3, 270, 480])
#   pts_seconds: tensor([0.0667, 3.3367], dtype=torch.float64)
#   duration_seconds: tensor([0.0334, 0.0334], dtype=torch.float64)

# Time-based indexing with PTS and duration info
decoder.get_frames_played_at(seconds=[0.5, 10.4])
# FrameBatch:
#   data (shape): torch.Size([2, 3, 270, 480])
#   pts_seconds: tensor([ 0.4671, 10.3770], dtype=torch.float64)
#   duration_seconds: tensor([0.0334, 0.0334], dtype=torch.float64)

Clip sampling

from torchcodec.samplers import clips_at_regular_timestamps

clips_at_regular_timestamps(
    decoder,
    seconds_between_clip_starts=1.5,
    num_frames_per_clip=4,
    seconds_between_frames=0.1
)
# FrameBatch:
#   data (shape): torch.Size([9, 4, 3, 270, 480])
#   pts_seconds: tensor([[ 0.0000,  0.0667,  0.1668,  0.2669],
#         [ 1.4681,  1.5682,  1.6683,  1.7684],
#         [ 2.9696,  3.0697,  3.1698,  3.2699],
#         ... (truncated), dtype=torch.float64)
#   duration_seconds: tensor([[0.0334, 0.0334, 0.0334, 0.0334],
#         [0.0334, 0.0334, 0.0334, 0.0334],
#         [0.0334, 0.0334, 0.0334, 0.0334],
#         ... (truncated), dtype=torch.float64)

You can use the following snippet to generate a video with FFmpeg and tryout TorchCodec:

fontfile=/usr/share/fonts/dejavu-sans-mono-fonts/DejaVuSansMono-Bold.ttf
output_video_file=/tmp/output_video.mp4

ffmpeg -f lavfi -i \
    color=size=640x400:duration=10:rate=25:color=blue \
    -vf "drawtext=fontfile=${fontfile}:fontsize=30:fontcolor=white:x=(w-text_w)/2:y=(h-text_h)/2:text='Frame %{frame_num}'" \
    ${output_video_file}

Installing TorchCodec

Installing CPU-only TorchCodec

1. Install the latest stable version of PyTorch following the official instructions. For other versions, refer to the table below for compatibility between versions of torch and torchcodec.

2. Install FFmpeg, if it's not already installed. Linux distributions usually come with FFmpeg pre-installed. TorchCodec supports major FFmpeg versions in [4, 7] on all platforms, and FFmpeg version 8 is supported on Mac and Linux.

   If FFmpeg is not already installed, or you need a more recent version, an easy way to install it is to use conda:

   conda install "ffmpeg<8"
   # or
   conda install "ffmpeg<8" -c conda-forge
   

3. Install TorchCodec:

   pip install torchcodec
   

The following table indicates the compatibility between versions of torchcodec, torch and Python.

torchcodec	torch	Python
main / nightly	main / nightly	>=3.10, <=3.13
0.8	2.9	>=3.10, <=3.13
0.7	2.8	>=3.9, <=3.13
0.6	2.8	>=3.9, <=3.13
0.5	2.7	>=3.9, <=3.13
0.4	2.7	>=3.9, <=3.13
0.3	2.7	>=3.9, <=3.13
0.2	2.6	>=3.9, <=3.13
0.1	2.5	>=3.9, <=3.12
0.0.3	2.4	>=3.8, <=3.12

Installing CUDA-enabled TorchCodec

First, make sure you have a GPU that has NVDEC hardware that can decode the format you want. Refer to Nvidia's GPU support matrix for more details here.

1. Install FFmpeg with NVDEC support. TorchCodec with CUDA should work with FFmpeg versions in [4, 7] on all platforms, and FFmpeg version 8 is supported on Linux.

   If FFmpeg is not already installed, or you need a more recent version, an easy way to install it is to use conda:

   conda install "ffmpeg<8"
   # or
   conda install "ffmpeg<8" -c conda-forge
   

   After installing FFmpeg make sure it has NVDEC support when you list the supported decoders:

   ffmpeg -decoders | grep -i nvidia
   # This should show a line like this:
   # V..... h264_cuvid           Nvidia CUVID H264 decoder (codec h264)
   

   To check that FFmpeg libraries work with NVDEC correctly you can decode a sample video:

   ffmpeg -hwaccel cuda -hwaccel_output_format cuda -i test/resources/nasa_13013.mp4 -f null -
   

Linux

2. Install Pytorch corresponding to your CUDA Toolkit using the official instructions. You'll need the libnpp and libnvrtc CUDA libraries, which are usually part of the CUDA Toolkit.

3. Install TorchCodec

   Pass in an --index-url parameter that corresponds to your CUDA Toolkit version, for example:

   # This corresponds to CUDA Toolkit version 12.6. It should be the same one
   # you used when you installed PyTorch (If you installed PyTorch with pip).
   pip install torchcodec --index-url=https://download.pytorch.org/whl/cu126
   

   Note that without passing in the --index-url parameter, pip installs the CPU-only version of TorchCodec.

Windows

2. On Windows (experimental support), you'll need to rely on conda to install both pytorch and TorchCodec:

   conda install -c conda-forge "torchcodec=*=*cuda*"
   

Benchmark Results

The following was generated by running our benchmark script on a lightly loaded 22-core machine with an Nvidia A100 with 5 NVDEC decoders.

The top row is a Mandelbrot video generated from FFmpeg that has a resolution of 1280x720 at 60 fps and is 120 seconds long. The bottom row is promotional video from NASA that has a resolution of 960x540 at 29.7 fps and is 206 seconds long. Both videos were encoded with libx264 and yuv420p pixel format. All decoders, except for TorchVision, used FFmpeg 6.1.2. TorchVision used FFmpeg 4.2.2.

For TorchCodec, the "approx" label means that it was using approximate mode for seeking.

Contributing

We welcome contributions to TorchCodec! Please see our contributing guide for more details.

License

TorchCodec is released under the BSD 3 license.

However, TorchCodec may be used with code not written by Meta which may be distributed under different licenses.

For example, if you build TorchCodec with ENABLE_CUDA=1 or use the CUDA-enabled release of torchcodec, please review CUDA's license here: Nvidia licenses.