RSS applies GRPO-style reinforcement learning with mIoU rewards to semantic segmentation and claims to beat weakly supervised methods, but the comparison is unfair and the results are unreproducible.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
Semantic segmentation with reward
RSS applies GRPO-style reinforcement learning with mIoU rewards to semantic segmentation and claims to beat weakly supervised methods, but the comparison is unfair and the results are unreproducible.