analiza aiqbench
This research addresses the granularity mismatch in reinforcement learning for Multimodal Large Language Models by introducing MCR-GRPO, which assigns precise box-level credit rather than broadcasting rewards across entire responses. By using a leave-one-out comparison to evaluate individual object contributions, the framework enables more stable optimization of structured multi-object grounding and segmentation tasks. The approach demonstrates state-of-the-art performance on benchmarks requiring visual recognition, language-to-object binding, and cardinality preservation.
Credit the Right Box: Marginal Contribution Assignment for Structured Visual Perception
arXiv:2608.01055v1 Announce Type: cross Abstract: Multimodal Large Language Models (MLLMs) are increasingly expected to solve structured perception tasks that require visual recognition, language-to-object binding, object cardinality preservation, and precisely localized grounding and segmentation outputs. However, existing group-relative reinforcement learning methods provide only response-level supervision, creating a granularity mismatch for structured multi-object prediction: a single advantage is broadcast to all tokens in a response, without distinguishing individual box contributions. To address this mismatch, we propose MCR-GRPO, a marginal contribution assignment framework that derives box-level credit directly from each sampled response. Specifically, Marginal Contribution Reward (MCR) estimates each predicted box's contribution through a leave-one-out comparison, measuring how the matched set value changes when the box is removed from the response. After within-response normalization, records that improve the set value receive positive credit, while redundant or harmful ones are suppressed. To make marginal attribution stable and informative, we further introduce a Continuous Matched Set Value Evaluator that integrates permutation-invariant matching, count-aware normalization, and graded localization. MCR-GRPO maps normalized box-level marginal advantages to the token spans that generated each box, preserving GRPO's response-level comparison while enabling box-aware optimization of structured multi-object grounding. Experiments across REC, DOD, segmentation, and counting benchmarks show state-of-the-art performance over prior GRPO-based baselines.
Abstract:Multimodal Large Language Models (MLLMs) are increasingly expected to solve structured perception tasks that require visual recognition, language-to-object binding, object cardinality preservation, and precisely localized grounding and segmentation outputs. However, existing group-relative reinforcement learning methods provide only response-level supervision, creating a granularity mismatch for structured multi-object prediction: a single advantage is broadcast to all tokens in a response, without distinguishing individual box contributions. To address this mismatch, we propose MCR-GRPO, a marginal contribution assignment framework that derives box-level credit directly from each sampled response. Specifically, Marginal Contribution Reward (MCR) estimates each predicted box's contribution through a leave-one-out comparison, measuring how the matched set value changes when the box is removed from the response. After within-response normalization, records that improve the set value receive positive credit, while redundant or harmful ones are suppressed. To make marginal attribution stable and informative, we further introduce a Continuous Matched Set Value Evaluator that integrates permutation-invariant matching, count-aware normalization, and graded localization. MCR-GRPO maps normalized box-level marginal advantages to the token spans that generated each box, preserving GRPO's response-level comparison while enabling box-aware optimization of structured multi-object grounding. Experiments across REC, DOD, segmentation, and counting benchmarks show state-of-the-art performance over prior GRPO-based baselines.
Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
arXiv:2608.01055 [cs.CV]
(or arXiv:2608.01055v1 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2608.01055 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 298 cuvinte
Subiecte conexe
OpenRouter: from Seed to Stripe — with OpenRouter’s Alex Atallah & AMP’s Anjney Midha
Latent Space · 25 sept. 2026
Links #5: 2026/07
LessWrong · 31 iul. 2026
P(Anyone Builds It & Everyone Dies) = 0.62?
LessWrong · 16 oct. 2026
Game Design Creativity evals
LessWrong · 12 oct. 2026