search

XinyiYing / BasicIRSTD

BasicIRSTD toolbox
132 stars 21 forks source link

Question for soft-IoU loss #24

Open xujiayi-hub opened 1 month ago

xujiayi-hub commented 1 month ago

In most referring papers for example UIUNet, they use BCE loss for optimization. Why you choose to use soft-IoU loss in most models？

XinyiYing commented 1 month ago

BCE loss exhibits inferior performance due to imbalance between positive and negative samples. Following ACM, ALCNet, DNANet, we employ soft-IoU loss. More discussion can be seen in 《Mapping Degeneration Meets Label Evolution: Learning Infrared Small Target Detection with Single Point Supervision》.