arXiv Computer Vision
Generalized CycleGRPO for Fine-Grained Multimodal Understanding and Grounding
Read the original on arXiv Computer Vision →The Flow has not summarised this story yet — read it at arXiv Computer Vision.
Related stories
arXiv Computer Vision
Sep 3
Actor as Its Own Critic: Unifying Region Understanding and Localization via CycleGRPO
The paper presents CycleGRPO, a reinforcement learning framework that unifies region understanding and localization for multimodal large language models (MLLMs). By treating the MLLM as both actor and critic, the method generates region captions and immediately grounds them back into spatial coordinates, using a token‑level cycle‑consistency reward that obviates the need for textual ground truths. Experiments on SAMTok demonstrate that CycleGRPO can bootstrap region captioning, VQA, grounded dialogue, and referring segmentation simultaneously, achieving consistent performance gains without task‑specific fine‑tuning.
arXiv AI
Oct 15
EventLens: Event-Structure Reinforcement Learning for Video Understanding
arXiv:2610.16172v1 Announce Type: cross Abstract: Recent multimodal large language models (MLLMs) achieve strong results on video question answering, yet often fail to recover the temporal structure...
arXiv AI
Aug 25
Think with Structured Grounding: Perceptual Reinforcement Learning for Chart and Visual-Tabular Understanding
arXiv:2608.22429v1 Announce Type: new Abstract: Multimodal Large Language Models (MLLMs) capable of thinking with images often rely on external tools for fine-grained perception. However, this relian...
arXiv AI
Oct 1
Reinforcing Multimodal Reasoning via Token-Level Perception-Grounded Advantage Estimation
arXiv:2609.39168v1 Announce Type: new Abstract: Reinforcement Learning with Verifiable Rewards (RLVR) has improved the reasoning capabilities of Multimodal Large Language Models (MLLMs), yet existing...
arXiv Computer Vision
Sep 16
SAVOR: Self-Aware Visual Grounding via Confidence-Calibrated Reinforcement Learning for Multimodal Hallucination Mitigation
SAVOR is a training framework for multimodal large language models that adds token and answer confidence to the output schema, optimises a Group Relative Policy Optimisation objective to penalise calibration error and poor abstention, and uses the learned confidence at inference to revisit visual evidence only when uncertain. Experiments on POPE, HallusionBench, AMBER, and MMHal-Bench with InternVL3-8B and Qwen3-VL-8B backbones show that SAVOR reduces hallucination while maintaining general capability on MME and MMBench, achieving lower Expected Calibration Error than DPO and decoding baselines.
arXiv Computer Vision
Oct 15
CASHEW: Stabilizing Multimodal Reasoning via Iterative Trajectory Aggregation
arXiv:2601.08010v4 Announce Type: replace Abstract: Vision-language models achieve strong performance across a wide range of multimodal understanding and reasoning tasks, yet their multi-step reasoni...