CompressAI

CompressAI is a PyTorch library and evaluation platform for end-to-end compression research. It provides custom operations, layers, models, and tools to research, develop, and evaluate image and video compression codecs. CompressAI includes pre-trained models and evaluation tools

Link

• Visit resource

Related resources

• MLVC: Multi-platform Learned Video Codec for Real-World DeploymentAcademic paper on a neural video codec targeting competitive compression, real-time speed, and cross-platform (Apple/Intel/Qualcomm NPU) in…
• Benchmarking Conventional and Learned Video Codecs with a Low-Delay ConfigurationAcademic benchmark comparing traditional (HEVC/VVC) and learned video codecs under low-delay configuration constraints.
• CompressAI-Vision: Video Compression for Machines BenchmarkPaper + framework for designing, testing and comparing "video coding for machines" pipelines, mixing neural CompressAI models against VVC/H…
• Cool-chic: Coordinate-based Low Complexity Hierarchical Image/Video CodecUltra-low-complexity overfitted/coordinate-based learned video codec using ~800 parameters, with companion open-source implementation.
• OpenDCVCs: Reproducible PyTorch Reimplementation of the DCVC Codec SeriesIndependent open reimplementation with training+eval code of the whole Microsoft DCVC learned video codec family (prior official releases w…
• Differentiable VMAF Re-implementation on PyTorchDifferentiable PyTorch reimplementation of VMAF metric enabling gradient-based codec/RDO optimization, with companion code.
• Performance Comparison of AV1, JEM, VP9, and HEVC EncodersFraunhofer HHI SPIE 2017 paper comparing codec internals, BD-BR, and complexity across AV1, JEM, VP9, and HEVC.
• nano-hevcMinimal educational HEVC encoder written in pure Python, readable-over-fast, for teaching codec internals.
• Opus (xiph/opus)Reference implementation of the Opus audio codec, recently adding deep-learning-based packet loss concealment.
• dav1d.wasmWASM port of the dav1d AV1 decoder, SIMD-optimized with TypeScript interfaces, aimed at production use in browsers.