TsallisPGD uses a tunable Tsallis cross-entropy loss with a dynamic q schedule to create stronger adversarial attacks that reduce accuracy and mIoU more effectively than prior methods on semantic segmentation tasks.
A survey on deep learning in medical image analysis,
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
TsallisPGD: Adaptive Gradient Weighting for Adversarial Attacks on Semantic Segmentation
TsallisPGD uses a tunable Tsallis cross-entropy loss with a dynamic q schedule to create stronger adversarial attacks that reduce accuracy and mIoU more effectively than prior methods on semantic segmentation tasks.