[ECCV 2026] Official code for CycleGRPO and its video extension Generalized CycleGRPO — unifying region understanding and localization via caption-grounding cycle-consistency RL
reinforcement-learning video-understanding dense-captioning visual-grounding cycle-consistency referring-expression-segmentation temporal-grounding vision-language-model sam2 grpo multimodal-large-language-model qwen3-vl eccv-2026 cyclegrpo region-captioning
-
Updated
Nov 12, 2026 - Python