Commit History

28 days ago

  1. Move v2 into CorridorKeyModule, drop licence files for links, Native quality, no frame cap, auto window by size Running

    1ef5ded
  2. Pin torch 2.9.1 (tested v2 version), hide UI-only event from API, document ZeroGPU size

    6dcf06b
  3. Add CorridorKey v2 (GPU) with v1/v2 switch, v2 default; v1 unchanged behind --version v1

    30aa4cc

29 days ago

  1. Add notice v1

    Verified signature
    309dd62

  1. postprocess at model res, defer resize+write to CPU (saves ~35s GPU)⁠

    f4a7288
  2. compact UI: resolution+screen color in row, despeckle in row

    18d73cb
  3. fix: remove tree paths from credit links

    51cee71
  4. pin upstream commit IDs in credits

    0cabf66
  5. fix: blue CPU is available (1024 only), not green-only

    119149d
  6. update all references: green → green/blue

    804c80b
  7. status shows CPU + GPU time breakdown

    323e65e
  8. add blue 1024 ONNX (FP16 on disk, FP32 at runtime), rename models⁠

    646f0cd
  9. blue screen support: GPU-only, auto-detection, no ONNX/LFS⁠

    7b2ad18
  10. safetensors loading, Phase 0 4x faster (uint8), total time in status⁠

    33e616b
  11. quality: lower clean_matte threshold 0.25→0.02, always keep largest component⁠

    1363975
  12. cleanup: stale comments, dead import, redundant makedirs, fix batch size in UI

    a2a7a3e
  13. simplify: merge write functions, fix missing Processed output, bulk transfer⁠

    9d23c67
  14. remove dead code: AOTI export, inductor/triton cache, shared_results, deferred write

    2a4471f
  15. disable torch.compile on ZeroGPU — net negative for GreenFormer⁠

    f4a2965
  16. fix: reduce-overhead instead of max-autotune (118s→~30s), dedicated export endpoint⁠

    c53eb28
  17. triple-fallback artifact export + triton cache⁠

    449e82f
  18. export .pt2 after compile + download button⁠

    7720399
  19. always compile with triton (max-autotune) + probe AOTI⁠

    4537006
  20. simplify pipeline timing in README

    cc14934
  21. enable triton+CUDA graphs via torch.compile(max-autotune)⁠

    f1d6e7f
  22. fix README: accurate torch.compile description, no triton/AOTI claim

    cdef1d9
  23. update README: fix batch size, add GPU optimization docs⁠

    69eef34
  24. fix randperm overflow at 4K: per-frame connected components

    c9b234a
  25. GPU postprocessing pipeline + TF32 + conditional torch.compile

    8ea0c8b

  1. add ZeroGPU GPU inference (FP16, flash-attn, batch=32@1024/16@2048)

    0b6961f

  1. professional output package (Comp+FG+Matte+Processed)

    c2d53e4
  2. download button for all output modes

    16d8be4
  3. fix gradio cache_examples None output crash

    d041804
  4. fix example cache crash

    40a8177
  5. initial commit

    20c856c

  1. initial commit

    2b1c753