Comparing the Robustness of ResNet, Swin-Transformer, and MLP-Mixer under Unique Distribution Shifts in Fundus Images
Kazuaki Ishihara, Koutarou Matsumoto, Hongqing Yu, Alaa AlZoubi, Yifan Zhao, Hongbo Du, Larbi Boubchir · 2023
Read Comparing the Robustness of ResNet, Swin-Transformer, and MLP-Mixer under Unique Distribution Shifts in Fundus Images by Kazuaki Ishihara, Koutarou Matsumoto, Hongqing Yu, Alaa AlZoubi, Yifan Zhao, Hongbo Du, Larbi Boubchir (2023) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/pmc-10740473.txt.
Description
Ishihara and co-authors stress-test three vision models on shifted eye images. Diabetic retinopathy is a top cause of vision loss. Models must hold up when clinic images differ from training sets. The team added fundus-specific noise to the EyePACS data set. ResNet, Swin-Transformer, and MLP-Mixer were compared head to head.
Swin-Transformer held the highest accuracy score under every noise type. Its score also fell least as noise grew. Calibration error showed no steady pattern across the models. The authors favor transformer designs for robust eye screening. Safety-critical use, they note, demands this kind of shift testing.
Catalog record
| CREATOR | Kazuaki Ishihara, Koutarou Matsumoto, Hongqing Yu, Alaa AlZoubi, Yifan Zhao, Hongbo Du, Larbi Boubchir |
|---|---|
| YEAR | 2023 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 29 KB |
| VERSION | 1.0 |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |