library_name: pytorch
license: other
license_name: polyform-noncommercial-1.0.0
license_link: LICENSE.md
pipeline_tag: image-feature-extraction
tags:
- endoscopy
- medical
- foundation-model
- gastric-cancer
GutCore model weights
GutCore is an endoscopy foundation model for whole-case gastric cancer analysis. See the medRxiv preprint and inference code.
[2026-09-01 Update] We have also trained a lightweight ViT-B variant that achieved comparable performance to ViT-L in our internal benchmarks. If you have an interest or use case for the ViT-B weights, please open a discussion/issue or contact us—we would be glad to release it if there is demand.
Files
| File | Contents |
|---|---|
GutCore-ViT-L.pth |
Pretrained image encoder state dictionary |
GutCore-GC.pth |
Fine-tuned encoder, gated attention aggregation, and cancer classifier state dictionary |
Each .pth file is a flat dict[str, Tensor] containing model tensors only.
Optimizer state, scheduler state, epochs, metrics, and training history are not
included. The JSON files record architecture and inference configuration.
SHA-256 digests are provided in checksums.sha256.
Official downloads are verified against the published SHA-256 checksums. Only
load a local .pth file when you trust its source.
Download
The GutCore package downloads the immutable v0.1.0 revision automatically:
uv run gutcore-download --model all
Files can also be downloaded directly:
hf download SehunFromDaegu/GutCore \
GutCore-ViT-L.pth GutCore-GC.pth \
--revision v0.1.0
Intended use
The models are released for noncommercial academic research. They are not
medical devices and must not be used for diagnosis, screening, prognosis,
treatment selection, or patient management. See MODEL_CARD.md, LICENSE.md,
and NOTICE.md before use or redistribution.