zhanghang1989/PyTorch-Encoding

Provides pretrained models and modules for semantic segmentation and image classification, including ResNeSt split-attention networks and context encoding layers optimized for dense prediction tasks. Built as a PyTorch library with custom CUDA kernels for efficient encoding operations, integrating with standard vision datasets (ImageNet, ADE20K, PASCAL Context). Implements research architectures from multiple papers with reproducible model zoo checkpoints.

2,049 stars. No commits in the last 6 months.