| Title |
Efficient vision transformer training routine for Diabetic Foot Ulcer semantic segmentation |
| Authors |
Kairys, Arturas ; Raudonis, Vidas |
| DOI |
10.1145/3787279.3787324 |
| ISBN |
9798400721045 |
| Full Text |
|
| Is Part of |
ICAAI 2025: proceedings of the 2025 9th international conference on Advances in Artificial Intelligence, 14-16 November 2026, Manchester, UK.. New York : ACM, 2025. p. 281-288.. ISBN 9798400721045 |
| Keywords [eng] |
Deep learning ; Diabetic foot ulcer ; SegFormer ; Semantic segmentation |
| Abstract [eng] |
Diabetic Foot Ulcer (DFU) segmentation plays a crucial role in automated wound assessment, aiding clini-cians in monitoring wound progression and guiding treatment decisions. However, lack of training data, inconsistent annotation or ambiguous scenery makes it difficult to finetune robust semantic segmentation models. In this study, we propose fine-tuning pipeline and use it to improve SegFormer segmentation re-sults. We explore different SegFormer sizes, dataset increase and alteration options, loss functions, preprocessing and postprocessing techniques. Our experiments demonstrate that applying proposed fine-tuning routine helps improve SegFormer segmentation results. Our final model achieved 0.733 F1 score and ranked 6th on the DFUC2022 leaderboard. |
| Published |
New York : ACM, 2025 |
| Type |
Conference paper |
| Language |
English |
| Publication date |
2025 |
| CC license |
|
