The authors examined confidence-based decoding in masked generative models, which predict many image tokens in parallel, retain the most confident predictions and remask the rest.
MaskSense preserves subject exploration when training masked image generators with RL
The framework changes both sampling and optimization so reinforcement learning can compare more diverse image-generation trajectories.
Chinese Tech
Yawen Shao · Jie Xiao · Kai Zhu · Yu Liu · Hongchen Luo · Xueyang Fu · +3 more
University of Science and Technology of China · Tongyi Lab, Alibaba Group · Northeastern University
Research Digest··2 min read
Shao and colleagues identify a mismatch between masked image generation and Group Relative Policy Optimization (GRPO), which learns by comparing rewards among groups of sampled outputs.
Why this paper
From Tongyi Lab, Alibaba Group and 2 others
In one line
MaskSense uses semantic priors and entropy anchoring to avoid a visual exploration trap in reinforcement learning for masked image generation.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ✓Reports numbers on named benchmarks (2 benchmarks)
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§