Skip to content

v1.0.0 — Liver restoration training and inference

Latest

Choose a tag to compare

@fengluo233 fengluo233 released this 22 Sep 06:36

Initial release of the AutolysisBench liver autolysis-restoration training and patch-inference code.

  • Pretrained final epoch-200 generator, MIT licensed.
  • Training preset, input-only five-step inference, English and Chinese documentation.
  • Verified seeded repeatability and five-step agreement with the research implementation in the tested environment.
  • Two-update training smoke test, all-network checkpoint reload and inference from newly trained weights passed.

Download A2F_10x_epoch200_net_G.pth and SHA256SUMS, then follow the repository README. MODEL_LICENSE.txt contains the MIT terms for the weights. The checkpoint SHA-256 is 812e4ad73b19bdc5e4b56c5351cc85bce7497834d621fba29398a4a864f38983.

Inputs are resized to 256 × 256. Blank/low-tissue patches may generate tissue-like textures; filter input tissue regions as described in the documentation. Raw data, whole-slide processing and other-organ experiments are not included.