Skip to content

CaC: Advancing Video Reward Models via Hierarchical Spatiotemporal Concentrating

May 2026 · arXiv.org · Vol abs/2605.11723 · 5 citations · 73 references
Computer Science

TL;DR

Concentrate and Concentrate (CaC) is a coarse-to-fine anomaly reward model based on Vision-Language Models that first conducts a global temporal scan to anchor anomalous time windows, then performs fine-grained spatial grounding within the localized interval, and finally derives robust judgments via structured spatiotemporal Chain-of-Thought reasoning.

Abstract

In this paper, we propose Concentrate and Concentrate (CaC), a coarse-to-fine anomaly reward model based on Vision-Language Models. During inference, it first conducts a global temporal scan to anchor anomalous time windows, then performs fine-grained spatial grounding within the localized interval, and finally derives robust judgments via structured spatiotemporal Chain-of-Thought reasoning. To equip the model with these capabilities, we construct the first large-scale generated video anomaly dataset with per-frame bounding-box annotations, temporal anomaly windows, and fine-grained attribution labels. Building on this dataset, we design a three-stage progressive training paradigm. The model initially learns spatial and temporal anchoring through single- and multi-frame supervised fine-tuning, and then is optimized by a reinforcement learning strategy based on two-turn Group Relative Policy Optimization (GRPO). Beyond conventional accuracy rewards, we introduce Temporal and Spatial IoU rewards to supervise the intermediate localization process, effectively guiding the model toward more grounded and interpretable spatiotemporal reasoning. Extensive experiments demonstrate that CaC can stably concentrate on subtle anomalies, achieving a 25.7% accuracy improvement on fine-grained anomaly benchmarks and, when used as a reward signal, CaC reduces generated-video anomalies by 11.7% while improving overall video quality.

View source

Similar papers

Sep 2026

Frame Differential On-Policy Self-Distillation for Video Reasoning

Reinforcement learning (RL) has substantially improved the reasoning ability of multimodal language models through verifiable rewards and increasingly fine-grainedvisual or temporal credit assignment. In video reasoning, however, current RL methods typically train with a fixed sparse frame budget: increasing the number...

Haiying He, Xin Zheng, Shao-Li Hu et al. · 0 citations
Preprint Oct 2026

Predictive Long-Horizon Progress Modeling with Action-Aware Quantification

Quantitative and dense task progress feedback is essential for robotic manipulation across data curation, policy evaluation, and reinforcement learning. Existing progress annotations often rely on sparse keyframe markings derived from subjective manual judgments, limiting their scalability and cross-task consistency, p...

Tian-Chi Xu, Ya-Lin Zheng, Yan Zhu et al. · 0 citations
Book Open access Nov 2026

Streaming-o1: Reinforcing Streaming Video Understanding with Adaptive Reasoning Depth

In recent years, multimodal large models have made significant progress in video understanding and reasoning. However, most of them rely on an offline setting, which assumes that both videos and questions can be fully accessed in advance. In contrast, streaming video question answering scenarios require capturing depen...

Shi-Han Wang, Zhen-Yu Yang, Li-Long Liu et al. · 0 citations
Preprint Oct 2026

ProAR: Learning Prospective Reasoning with Autoregressive Video Models

Autoregressive (AR) video models excel at causal generation, but their reliance on next-chunk prediction confines them to a short-sighted, reactive paradigm. This limitation is particularly consequential for reasoning-oriented generation, where achieving a target outcome through valid intermediate states matters more t...

Ling-Hui Shen, Ting-Hui Zhu, Sheng Zhang et al. · 0 citations
#artificial intelligence Review Oct 2026

Getting Space Right: Agentic Rewards for Image Generation

Reinforcement learning (RL) is promising for spatial text-to-image generation, but existing work largely focuses on policy optimization, assuming accurate rewards and effective online rollouts. We observe in-domain gains with out-of-domain degradation, exposing two limitations: unreliable spatial rewards in open-world...

Xiaofeng Tan, Jia-Xin Lin, Haobo Yuan et al. · 1 citation

Related blog posts

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.