MPCompress v0.1.0

MPCompress v0.1.0

MPCompress is a PyTorch library and evaluation platform for multi-purpose compression research. This is the first stable release with a unified framework, comprehensive API documentation, and evaluation tools.

🎯 Core Framework

• Multi-layer Representation Video Coding Framework: Introduced a flexible coding architecture with unified interfaces for Data Unit (DU), Frame, and Video codecs
• Unified Codec Interfaces: Standardized DataUnitCodec, FrameCodec, and VideoCodec interfaces with support for FrameWise and LayerWise organization
• Intermediate Representation: Defined coded_unit and coded_data formats for seamless integration between compression and decompression pipelines

🚀 Model Support

• MPC (Multi-Purpose Compression): Refactored to API v0.1, prioritizing machine vision while maintaining human visual compatibility
• LaMoFC (Large Model Feature Coding): Initial implementation for distributed large model deployments with feature compression
• DCVC-RT: Real-time video feature compression reference implementation with I-frame/P-frame support

🛠️ Training & Evaluation

• Offline Training Pipeline: Added offline feature extraction and training scripts for efficient experimentation
• Unified Dataset Interface: Standardized image and video dataset formats with consistent metadata structures

📚 Documentation

• Complete API Documentation: Added comprehensive docstrings across all modules
• Documentation Site: Built with Zensical/MkDocs and deployed to GitHub Pages

🔧 Utilities & Tools

• Stream Helpers: High-level syntax reading/writing utilities for bitstream manipulation
• Debug Tools: Enhanced debugging tools with nested structure shape extraction and tensor hash calculation