Image Cutout — model provenance and attribution
================================================

Model: SlimSAM-77 (uniform) — Zigeng Chen, Gongfan Fang, Xinyin Ma, Xinchao Wang,
"SlimSAM: 0.1% Data Makes Segment Anything Slim" (NeurIPS 2024).
Upstream code: https://github.com/czg1225/SlimSAM — Apache License 2.0.
Derived from Segment Anything (Kirillov et al., Meta AI) — Apache License 2.0.

ONNX export used unmodified: https://huggingface.co/Xenova/slimsam-77-uniform
Pinned revision: 5850ab45f587c112167512ffef949107115e26a0 (repo card licence: apache-2.0)

Files served from https://models.skillsafe.ai/ (SHA-256 of the exact bytes):
  onnx/vision_encoder_fp16.onnx                 12170657 bytes  11aaeb49c75e7b3f4cbf8a32c2c819406520c6b3affb4068ff474b2240c8aa38
  onnx/prompt_encoder_mask_decoder_fp16.onnx     8550118 bytes  df24d49a6f1a5dc0dbbecd84ca0fff9f14c76e63b81fd35c2b92c1321b007f71
  onnx/vision_encoder_quantized.onnx             8882165 bytes  cce23c7b2e5d4f330932738fb67ba518e04b0d99ccdd1cccd22a7da4e01f2971
  onnx/prompt_encoder_mask_decoder_quantized.onnx 4903810 bytes  cb90b279f549d2cab7fd6e20c38522438c65d84bdcca3d2a764cff7d857fdce2

The app verifies each download against these hashes before use.
Licence text: /models/LICENSE-Apache-2.0.txt. Platform attribution page: /models.txt.

Runtime: onnxruntime-web 1.27.0 (Microsoft, MIT) — dist/ort.bundle.min.mjs and
dist/ort-wasm-simd-threaded.jsep.wasm, vendored byte-for-byte from the npm package.

------------------------------------------------------------------------------------------------

Model: SAM ViT-Base — Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland,
Laura Gustafson, Tete Xiao, Spencer Whitehead, Alexander C. Berg, Wan-Yen Lo, Piotr Dollár,
Ross Girshick, "Segment Anything" (ICCV 2023), Meta AI — Apache License 2.0.
Upstream code: https://github.com/facebookresearch/segment-anything

ONNX export used unmodified: https://huggingface.co/Xenova/sam-vit-base
Pinned revision: e9ff25bcbb89d0be0758fcdf28e91193bb7eb6e9 (repo card licence: apache-2.0)

Files served from https://models.skillsafe.ai/ (SHA-256 of the exact bytes):
  onnx/vision_encoder_fp16.onnx                180194619 bytes  806b471147ba0b6e8dc33f912aecb35bd29c1eb337b6a7acb0bf9bed0766cc8b
  onnx/vision_encoder_quantized.onnx           101088469 bytes  d9d7bca3b256ab71b3b7cdc35839983bc8ebaf68ea9022f15805ac43955cd247
  onnx/prompt_encoder_mask_decoder_fp16.onnx     8550118 bytes  df24d49a6f1a5dc0dbbecd84ca0fff9f14c76e63b81fd35c2b92c1321b007f71  (identical to SlimSAM-77's)
  onnx/prompt_encoder_mask_decoder_quantized.onnx 4903810 bytes  cb90b279f549d2cab7fd6e20c38522438c65d84bdcca3d2a764cff7d857fdce2  (identical to SlimSAM-77's)

SlimSAM prunes the image encoder only; its prompt encoder + mask decoder are Meta's original
weights, which is why the decoder files above are byte-identical and downloaded once.