decord2
Decord2 is a high-performance, efficient video decoding and loading library for deep learning research, featuring smart shuffling, random frame access, GPU acceleration, and seamless integration with popular frameworks.
Decision gist · record as of 2026-08-14
Yes, with conditions. Install if you need efficient random video frame access for deep learning and can tolerate medium setup friction. Prebuilt wheels work out-of-the-box for CPU use on recent Python versions, but GPU acceleration requires source compilation. Active maintenance, no known vulnerabilities, and permissive licensing make it safe to adopt. Skip if your workload is sequential video playback or if building from source is not feasible.AI-flagged interpretation of the facts on this page — verify before relying
Before you install
- PYPI wheels are CPU-only; GPU acceleration (NVIDIA NVDEC, Intel codecs) requires building from source with appropriate compiler and CUDA toolkit.
- Requires Python >=3.9.0.
- Medium install friction: prebuilt wheels available for recent Python versions (3.10–3.14) on Linux, macOS, and Windows, but GPU acceleration requires building from source with CUDA/compiler toolchain.
License · maintenance · safety
Apache-2.0 (permissive) — Apache-2.0 permissive license allows commercial and private use with minimal restrictions; attribution required.
last release 2026-05-30 (76 days) · last repo commit 2026-05-30 · 64 stars
0 known vulnerabilities (OSV.dev, 2026-08-14) · 613,076 downloads/mo, #5,753 on PyPI
Alternatives
Verify before relying
pip install decord2
from decord import VideoReader, cpu
vr = VideoReader('video.mp4', ctx=cpu(0))
frames = vr.get_batch([0, 5, 10])
print(frames.shape)- Whether Intel codec support is production-ready or still in development
- Performance benchmarks against other video loaders for typical deep learning workloads
- Audio-video synchronization accuracy and latency characteristics
- Memory overhead of batch loading and internal frame caching strategies
What it is and what it does
Decord2 is a video and audio decoding library optimized for deep learning pipelines. It wraps hardware-accelerated decoders (FFMPEG, NVIDIA NVDEC, Intel codecs) to provide efficient random frame access and batch loading—operations that are typically slow in standard video libraries. The library handles the awkward seeking patterns common during neural network training by using smart shuffling and caching strategies to minimize redundant decoding.
The package provides three main interfaces: VideoReader for direct frame access from files, VideoLoader for training-scale batch loading across many video files, and AudioReader for audio extraction. It also offers AVReader to decode video and audio simultaneously with frame-level synchronization. Runtime dependency is numpy; GPU acceleration requires building from source with CUDA.
Use it for
- Load random frames from videos during deep learning training without expensive seeks or redundant decoding.
- Batch-load frames from multiple video files with configurable shuffling strategies for efficient data pipeline.
- Extract synchronized audio and video samples from media files for multimodal model training.
- Decode audio from video files at custom sample rates and channel layouts for audio processing tasks.
- Stream video frames in-memory from file-like objects for cloud or streaming scenarios.
Worth the install?
AI-flagged interpretation of the facts on this page. Verify before relying on it.
Yes, with conditions.
Install if you need efficient random video frame access for deep learning and can tolerate medium setup friction. Prebuilt wheels work out-of-the-box for CPU use on recent Python versions, but GPU acceleration requires source compilation. Active maintenance, no known vulnerabilities, and permissive licensing make it safe to adopt. Skip if your workload is sequential video playback or if building from source is not feasible.
Install
decord2 on PyPI
Before you install
Medium install friction: prebuilt wheels available for recent Python versions (3.10–3.14) on Linux, macOS, and Windows, but GPU acceleration requires building from source with CUDA/compiler toolchain. Active maintenance with recent commits.
PYPI wheels are CPU-only; GPU acceleration (NVIDIA NVDEC, Intel codecs) requires building from source with appropriate compiler and CUDA toolkit. Requires Python >=3.9.0.
License in practice
Apache-2.0 permissive license allows commercial and private use with minimal restrictions; attribution required.
Quickstart
pip install decord2
from decord import VideoReader, cpu
vr = VideoReader('video.mp4', ctx=cpu(0))
frames = vr.get_batch([0, 5, 10])
print(frames.shape)Verify before relying
- Whether Intel codec support is production-ready or still in development
- Performance benchmarks against other video loaders for typical deep learning workloads
- Audio-video synchronization accuracy and latency characteristics
- Memory overhead of batch loading and internal frame caching strategies
Package facts
| License | Apache-2.0 permissive |
| Python support | Supports the current Python release >=3.9.0 |
| Install friction | Medium. Platform-specific wheel |
| Runtime dependencies | 1 packagenumpy |
| Maintenance | Actively maintained 76 days since the last release |
| Last repo commit | |
| First released | |
| Downloads | 613,076 / month, #5,753 on PyPI 30-day window, as of 2026-08-14 |
| Known vulnerabilities | None known OSV.dev, checked 2026-08-14 |
Evidence: decord2-3.4.0-cp310-cp310-macosx_14_0_arm64.whl; decord2-3.4.0-cp310-cp310-manylinux_2_28_aarch64.whl; decord2-3.4.0-cp310-cp310-manylinux_2_28_x86_64.whl; decord2-3.4.0-cp310-cp310-win_amd64.whl; decord2-3.4.0-cp311-cp311-macosx_14_0_arm64.whl; decord2-3.4.0-cp311-cp311-manylinux_2_28_aarch64.whl; decord2-3.4.0-cp311-cp311-manylinux_2_28_x86_64.whl; decord2-3.4.0-cp311-cp311-win_amd64.whl; decord2-3.4.0-cp312-cp312-macosx_14_0_arm64.whl; decord2-3.4.0-cp312-cp312-manylinux_2_28_aarch64.whl; decord2-3.4.0-cp312-cp312-manylinux_2_28_x86_64.whl; decord2-3.4.0-cp312-cp312-win_amd64.whl; decord2-3.4.0-cp313-cp313-macosx_14_0_arm64.whl; decord2-3.4.0-cp313-cp313-manylinux_2_28_aarch64.whl; decord2-3.4.0-cp313-cp313-manylinux_2_28_x86_64.whl; decord2-3.4.0-cp313-cp313-win_amd64.whl; decord2-3.4.0-cp314-cp314-macosx_14_0_arm64.whl; decord2-3.4.0-cp314-cp314-manylinux_2_28_aarch64.whl; decord2-3.4.0-cp314-cp314-manylinux_2_28_x86_64.whl; decord2-3.4.0-cp314-cp314t-macosx_14_0_arm64.whl
Tags
Let your AI agent find packages like this
Example. Real query, live index.
You found this page by searching. An agent finds it by wishing: SkillFed indexes 14,416 PyPI packages by what they can do, searchable in plain language.
wish › “video decoding for deep learning”
- decord2Decord2 decodes video and audio files with hardware acceleration,…
- decordDecord decodes video and audio files with hardware-accelerated…
- transparent-backgroundRemoves backgrounds from images and videos using a deep learning…
Give your agent the search over MCP, or paste the wish link into any chat.
More Artificial Intelligence packages
LiteLLM provides a unified Python interface to call 100+ LLM providers (OpenAI, Anthropic, Gemini, Bedrock, Azure, and others) using OpenAI-compatible API format, available as both a Python SDK and a self-hosted AI Gateway proxy server.
Install it if you need to work with multiple LLM providers or want to centralize LLM routing in your organization.
Client library and CLI tool for downloading, uploading, and managing models, datasets, and repositories on the Hugging Face Hub platform.
Install it if you work with Hugging Face Hub models or datasets.
LangChain provides a framework for building agents and LLM-powered applications by composing language models, tools, and memory through a unified API that abstracts over multiple model providers.
hf-xet provides chunk-based deduplication and efficient file transfer for the Hugging Face Hub, enabling faster uploads and downloads of large files with local disk caching.
Tokenizers converts raw text into token sequences for NLP models, with support for training custom vocabularies and using pre-built tokenizers (BPE, WordPiece) optimized for speed via Rust.
Transformers provides a unified framework for loading, fine-tuning, and running state-of-the-art pretrained models across text, vision, audio, video, and multimodal tasks using PyTorch, JAX, or TensorFlow.
Install it if you need to run or train any transformer-based model for NLP, vision, audio, or multimodal tasks.
See also decord · torchcodec · nvidia-nvimgcodec-cu12 · imageio-ffmpeg · pynvvideocodec · audioread · moviepy · scikit-video · torch · nvidia-libnvcomp-cu12