Paper

ViRA: Toward Ultra-High-Resolution Visual Agents via Cross-Input Relative Policy Optimization · ByteTrail