Skip to main content

๐Ÿงญ DA-SAM3

A parameter-efficient SAM 3 medical adapter using a Dynamic Expert Router (DER) and Decomposed Parameterized Experts (DPE) to route text/visual features through lightweight experts.

Official codePaperWeights

โ† Model atlas

At a glanceโ€‹

FieldDetails
Model typeParameter-efficient medical SAM3 adaptation
Research team / organizationShenzhen Research Institute, The Chinese University of Hong Kong, Shenzhen; University of Science and Technology of China; Nanyang Technological University; Imperial College London (Imperial Global Singapore).
Input / modalityMedical image segmentation; the public repository lists Synapse, MMWHS, BTCV and ACDC datasets.
Prompt / interactionNatural-language text prompt combined with image-conditioned expert routing.
Code licenseThe GitHub README says research use only and the repository has no top-level LICENSE in the checked snapshot. SAM 3 base weights remain subject to Meta's SAM License. The MICCAI paper page lists no code URL, while the public GitHub README calls itself an official implementation.

๐Ÿ“š Paper and publicationโ€‹

Dual-Adaptive SAM3: Hierarchical Routing over Low-Rank Expert Layers for Parameter-Efficient Medical Image Segmentation โ€” MICCAI 2026, LNCS 16884, pp. 155โ€“164; arXiv:2607.02571.

๐Ÿ“ฆ Checkpoints and accessโ€‹

The project links the gated SAM 3 base checkpoint, but does not list a downloadable DA-SAM3 checkpoint. Its inference example expects a locally produced outputs/da_sam3/best_warmup.pt.

The code, model checkpoint and any required foundation-model checkpoint can have different terms. Review each linked source before use or redistribution.

โš™๏ธ Official setup and inferenceโ€‹

git clone https://github.com/Reconsider80/DA-SAM3.git
cd DA-SAM3
git clone https://github.com/Chongcong/Medical-SAM3.git ../Medical-SAM3
pip install -r requirements.txt
pip install -e .
# Follow the README to set SAM3_ROOT, PYTHONPATH and HF_TOKEN.
python train.py --config configs/da_sam3_default.yaml --stage warmup
# After training / obtaining a compatible DA checkpoint:
python infer.py --config configs/da_sam3_default.yaml \
--checkpoint outputs/da_sam3/best_warmup.pt \
--image path/to/ct_slice.png --prompt "left ventricle" --output result.png

โš ๏ธ Version and practical notesโ€‹

The public implementation does not list a downloadable DA-SAM3 checkpoint. Record the Git revision, SAM 3 base checkpoint, training stage and adapter checkpoint; the paper page and GitHub README currently describe code availability differently.

๐Ÿ“œ Licenseโ€‹

The GitHub README says research use only and the repository has no top-level LICENSE in the checked snapshot. SAM 3 base weights remain subject to Meta's SAM License. The MICCAI paper page lists no code URL, while the public GitHub README calls itself an official implementation.

Citation: Cite the paper above and the upstream project as directed by its authors. This repository links to the original paper; it does not redistribute the code or weights.

Source pages checked: 27 October 2026. Upstream environments, weights and access policies may change; use the linked project as the source of truth.