thomaschuwm/photo-demosaic-models · 摄影工具箱去马赛克模型分发仓

Photography toolbox demosaic model distribution repo — Thomas(储卫民)摄影工具箱(tksj)在役/评估通过的去马赛克模型权重公开分发。非商用项目;各模型许可与上游传导义务见各自目录的 LICENSE / NOTICE 与模型卡。

Distribution of demosaic model weights used by the Thomas photography toolbox. Non-commercial project; per-model licenses and upstream attribution obligations live in each model directory (LICENSE / NOTICE / model card).

模型索引 / Model Index

模型 / Model 目录 / Path 许可 / License 大小 / Size 简述
alcedo-bayer-s24-d8 models/alcedo-bayer-s24-d8/ GPL-3.0-only 155.1 KiB Alcedo Bayer 去马赛克蒸馏学生模型
alcedo-xtrans-p2-s32-d4 models/alcedo-xtrans-p2-s32-d4/ GPL-3.0-only 133.3 KiB Alcedo X-Trans 去马赛克蒸馏学生模型
mmnet-bayer-noisy models/mmnet-bayer-noisy/ MIT 2.78 MiB MMNet Bayer 去马赛克+降噪联合模型

布局 / Layout

models/<model-id>/
  <model-id>.<ext>     # 权重(bytes/sha256 见同目录 manifest.json / weights pinned by manifest.json)
  README.md            # 模型卡(中英双语 / bilingual model card)
  LICENSE              # 许可全文(按上游许可传导 / upstream license text)
  NOTICE               # 署名与第三方声明(如上游要求保留 / when upstream requires preservation)
  manifest.json        # 字节数/SHA-256/来源/评估口径(bytes/sha256/source/eval numbers)

完整性 / Integrity

每个权重文件的字节与 SHA-256 钉值登记在站点仓 src/lib/photoAnalysis/modelStaging.ts(分发源头账本)与本仓各 manifest.json,两处一致方可上传(上传脚本 fail-closed 校验)。 Every weight file's byte count and SHA-256 pin is registered in the site repo's modelStaging.ts (source-of-truth ledger) and mirrored in each manifest.json; the upload script verifies both and fails closed on any mismatch.

上游与署名 / Upstream & Attribution

  • Alcedo 学生权重:AlcedoStudio(GPL-3.0-only,v0.3.0)原件再分发,许可全文与 NOTICE 随目录分发。 Alcedo student weights are verbatim redistribution of AlcedoStudio v0.3.0 artifacts under GPL-3.0-only, with LICENSE and NOTICE preserved.
  • MMNet ONNX:cig-skoltech/deep_demosaick(MIT)上游 .pth 的本仓 ONNX 转换衍生品(转换脚本与对拍门见站点仓 tools/convert-mmnet-dmosaic.py)。 MMNet ONNX is a conversion derivative (MIT) of the upstream .pth, converted by the site repo's tooling with a numerical parity gate.

由站点仓备料批 pa-hfprep-1004 的 scripts/hf-upload.mjs 确定性生成(2026-10-05 04:30:50 UTC);凭证零硬编码,上传经 HF_TOKEN 环境变量注入。

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support