Learning Spatiotemporal Sensitivity in Video LLMs via Counterfactual Reinforcement Learning

arXiv:2605.21988 · cs.CV, cs.AI · Submitted 2026-05-21 · Read on arXiv

cs.CV, cs.AI

Submitted: 2026-05-21

Updated: 2026-09-27

Code: https://github.com/open-compass/VLMEvalKit

Project page: https://ddz16.github.io/crpo.github.io

Terminology

• GRPO — 55× in this paper · explained in 6 episodes
• Benchmark — 43× in this paper · explained in 2 episodes
• Accuracy — 40× in this paper · explained in 1 episode
• Counterfactual Relation Reward — 9× in this paper
• Reinforcement Learning — 8× in this paper · explained in 4 episodes
• Data Augmentation — 5× in this paper · explained in 7 episodes
• Counterfactual Relational Policy Optimization — 4× in this paper
• Ablation — 3× in this paper · explained in 1 episode
• Adam — 3× in this paper · explained in 2 episodes
• Fine-Tuning — 3× in this paper · explained in 5 episodes
• Hallucination — 3× in this paper · explained in 3 episodes
• Benchmarking — 2× in this paper · explained in 3 episodes
• Grounding — 2× in this paper · explained in 3 episodes
• Instruction Tuning — 2× in this paper · explained in 2 episodes
• Reward Function — 2× in this paper · explained in 2 episodes
• Temporal Reasoning — 2× in this paper · explained in 2 episodes
• Trace — 2× in this paper · explained in 2 episodes
• Ablation study — 1× in this paper · explained in 1 episode
• Actor — 1× in this paper · explained in 3 episodes
• Consistency — 1× in this paper · explained in 3 episodes
• Core — 1× in this paper · explained in 2 episodes
• Credit assignment — 1× in this paper · explained in 3 episodes
• Fidelity — 1× in this paper · explained in 2 episodes
• Generalization — 1× in this paper · explained in 13 episodes
• Multimodal Reasoning — 1× in this paper · explained in 5 episodes
• Reward Hacking — 1× in this paper · explained in 8 episodes
• Reward Shaping — 1× in this paper · explained in 2 episodes
• Verification — 1× in this paper · explained in 2 episodes

Sources

• STRIVE: Structured Spatiotemporal Exploration for Reinforcement Learning in Video Question Answering
• Qwen3-VL Technical Report
• Qwen2.5-VL Technical Report
• TemporalBench: Benchmarking Fine-grained Temporal Understanding for Multimodal Video Models
• PerceptionLM: Open-Access Data and Models for Detailed Visual Understanding
• Molmo2: Open Weights and Data for Vision-Language Models with Video Understanding and Grounding
• Lost in Time: A New Temporal Benchmark for VideoLLMs
• OneThinker: All-in-one Reasoning Model for Image and Video
• REINFORCE++: Stabilizing Critic-Free Policy Optimization with Global Advantage Normalization
• A Shortcut-aware Video-QA Benchmark for Physical Understanding via Minimal Video Pairs
• TimeBlind: A Spatio-Temporal Compositionality Benchmark for Video LLMs
• LLaVA-OneVision: Easy Visual Task Transfer
• VideoChat-R1: Enhancing Spatio-Temporal Perception via Reinforcement Fine-Tuning
• VideoAuto-R1: Video Auto Reasoning via Thinking Once, Answering Twice
• Part I: Tricks or Traps? A Deep Dive into RL for LLM Reasoning
• LongVU: Spatiotemporal Adaptive Compression for Long Video-Language Understanding
• VideoLoom: A Video Large Language Model for Joint Spatial-Temporal Understanding
• OpenAI GPT-5 System Card
• VL-Rethinker: Incentivizing Self-Reflection of Vision-Language Models with Reinforcement Learning
• InternVL3.5: Advancing Open-Source Multimodal Models in Versatility, Reasoning, and Efficiency

Related papers

• Loss Knows Best: Detecting Annotation Errors in Videos via Loss Trajectories
• AnchorWeave: World-Consistent Video Generation with Retrieved Local Spatial Memories
• Benchmarking the Robustness of Foundation Models for Mammography under Domain Shift
• MambaX-Net: Dual-Input Mamba-Enhanced Cross-Attention Network for Longitudinal MRI Segmentation
• TeleOCR: Navigating Document Parsing Across Digital and Camera-Captured Documents
• A Survey on Efficient Vision-Language-Action Models