Code repository for our paper titled "Segment Anything Is Not Always Perfect: An Investigation of SAM on Different Real-World Applications" (CVPRW Oral).
- Another work, Medical SAM Adapter which addresses the issue of lacking domain-specific medical knowledge in the SAM, are available now.
- Long version of this work has been accepted by Machine Intelligence Research.
- This work is awarded as Best Paper (Most Insightful Paper) at the CVPR'23 VISION Workshop.
- Evaluation code has been released.
- This work has been accepted as an Oral Presentation at the CVPR'23 VISION Workshop.
- Download the vit_b, vit_h and vim_l model from https://github.com/facebookresearch/segment-anything then put these models to the model_ck folder.
- Prepared own datasets put into the datasets folder.
- Set right path in /scripts/amg.py, then:
run amg.py
- After inferring, the SAM model generates predicted maps from a singer RGB image (multimask_output=True). Check right path in sam_dice_f1_mae.py or sam_f1_dice_mae.py to decide the best map selected by Dice or F1 metrics.
- Prepared these methods predicted maps to put into the other_methods_output folder.
- Check right path in /scripts/other_methods_dice_mae.py, then:
run other_methods_dice_mae.py
The download links of the dataset involved in our work are provided below.
DUTS | COME15K | VT1000 | DIS | COD10K | SBU | CDS2K | ColonDB |
---|---|---|---|---|---|---|---|
Link | Link | Link | Link | Link | Link | Link | Link |
If you find our work useful for your research or applications, please cite using this BibTeX:
@misc{ji2023segment,
title={Segment Anything Is Not Always Perfect: An Investigation of SAM on Different Real-world Applications},
author={Wei Ji and Jingjing Li and Qi Bi and Tingwei Liu and Wenbo Li and Li Cheng},
year={2023},
eprint={2304.05750},
archivePrefix={arXiv},
primaryClass={cs.CV}
}
@misc{wu2023medical,
title={Medical SAM Adapter: Adapting Segment Anything Model for Medical Image Segmentation},
author={Junde Wu and Wei Ji and Yuanpei Liu and Huazhu Fu and Min Xu and Yanwu Xu and Yueming Jin},
year={2023},
eprint={2304.12620},
archivePrefix={arXiv},
primaryClass={cs.CV}
}
Thanks for the efforts of the authors involved in the Segment Anything.