Segment Anything Is Not Always Perfect

Code repository for our paper titled "Segment Anything Is Not Always Perfect: An Investigation of SAM on Different Real-World Applications" (CVPRW Oral).

Updates

• [x] Another work, Medical SAM Adapter which addresses the issue of lacking domain-specific medical knowledge in the SAM, are available now.
• [x] Long version of this work has been accepted by Machine Intelligence Research.
• [x] This work is awarded as Best Paper (Most Insightful Paper) at the CVPR'23 VISION Workshop.
• [x] Evaluation code has been released.
• [x] This work has been accepted as an Oral Presentation at the CVPR'23 VISION Workshop.

Get Started

Eval SAM in different dataset

1. 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.
2. Prepared own datasets put into the datasets folder.
3. Set right path in /scripts/amg.py, then:

run amg.py

Chosen best results form the sam_output folder

1. 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.

Eval other methods in different dataset

1. Prepared these methods predicted maps to put into the other_methods_output folder.
2. Check right path in /scripts/other_methods_dice_mae.py, then:

run other_methods_dice_mae.py

Datasets

The download links of the dataset involved in our work are provided below.

Citation

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}
}

Acknowledgement

Thanks for the efforts of the authors involved in the Segment Anything.