Vision-RL2: Region-Level RL Matches Full-Resolution MLLM Accuracy With 4x Fewer Visual Tokens
Yuheng Shi · hf · 2026-09-18
Fine-grained perception in MLLMs is usually improved by raising resolution, but extra visual tokens inflate encoding and prefilling costs. The authors show via controlled diagnostics that localizing a region of interest tolerates roughly 3–4x stronger token compression than recognizing its content.
They propose Vision-RL2: region-level reinforcement learning optimizes a lightweight proposal network distilled from the model's attention. Coherent regions are treated as actions, and a frozen MLLM reader scores each by how its removal changes answer likelihood. Complementary subtractive and additive objectives suppress distracting proposals and recover missing evidence — no region annotations, response sampling, or reasoning trajectories needed. The refined proposal enables sparse encoding that magnifies evidence and excludes background tokens.
Across six fine-grained benchmarks and four MLLM backbones, Vision-RL2 beats the base model at every token budget and surpasses its largest-budget accuracy with about 4x fewer visual tokens. Code is available on GitHub.
More from Research
- Nielsen clarifies: arithmetic mixtures are normalized by default, geometric ones are not — FrnkNlsn · 2026-09-18
- Frank Nielsen publishes paper on two types of geometric Jensen–Shannon divergences — FrnkNlsn · 2026-09-18
- Pure-Rust visual SLAM library visloc-rs benchmarks 3.46x faster than COLMAP on ETH3D — rsasaki0109 · 2026-09-18
- Goodfire's activation monitors catch reward hacking in real time — models do it in 50-96% of rollouts — soleio · 2026-09-18
- Anthropic publishes three transparency metrics: AI-driven R&D, agent oversight, compute allocation — repligate · 2026-09-18
- Periodic Labs details its stack: 4.1x Megatron throughput, frontier-beating science models on 1,300 H200s — hsu_byron · 2026-09-18