# dmlc/decord

An efficient video loader for deep learning with smart shuffling that's super easy to digest

Repository: https://github.com/dmlc/decord
Canonical: https://ross.abutalabs.com/products/decord
Language: C++
License: Apache-2.0
License Family: permissive
Topics: video-loader
Last push: 2024-07-17T04:18:40+00:00

## Health v2 (maintenance only)
Score: 23/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 0, release rhythm 8, longevity 100
- inputs: {"age_days": 2784, "days_push": 777, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 2514, forks 233 (observed 2026-08-28T04:06:57.592104+00:00)

## What it is
Decord is a C++ library with Python bindings that provides efficient video and audio decoding for deep learning pipelines. It wraps hardware-accelerated decoders like FFMPEG/LibAV and Nvidia NVDEC and specializes in fast random access and shuffling of video frames during training.

## Use cases
- load video frames randomly for training neural networks
- decode video datasets with random access patterns
- slice video and audio in sync for multimodal models
- build video classification or action recognition data loaders
- accelerate video decoding with GPU (NVDEC)
- extract audio tracks from video files for ML

## When to choose
- you need fast random frame access from videos during deep learning training
- you want synchronized video and audio decoding in one tool
- you need hardware-accelerated decoding via NVDEC or FFmpeg backends
- you want a simple pip-installable video loader with framework bridges

## When to avoid
- you only need video playback or editing rather than ML data loading
- you need actively developed features or recent codec support
- you require GPU decoding via prebuilt pip wheels (only CPU wheels are published)
- you need a full video processing/transcoding toolkit

## Facets
- artifact type: library
- maturity: maintenance
- function: video-processing, audio-processing, machine-learning, data-science
- domain: machine-learning, deep-learning, computer-vision, media
- platform: windows, python, cpp
- tags: video-loader, video-decoding, random-access, ffmpeg, nvidia-codecs, data-loading, pytorch, tensorflow, video, linux, macos, gpu

## Member repositories
- dmlc/decord (main) score 23

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:06:57.592104+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T02:26:13.587394+00:00, confidence not recorded.
  - readme: https://github.com/dmlc/decord (fetched 2026-08-28T04:06:57.592104+00:00, sha 8489f06dd727)
- Data as of 2026-08-30T08:39:29.467469+00:00.