Video R1 Eval: Evaluation Dataset for Reinforcing Video Reasoning in MLLMs

Evaluation benchmarks for the Video-R1 model across video reasoning categories, including test sets for temporal and causal logic. The dataset provides the data required to replicate the reasoning performance results presented in the 'Video-R1: Reinforcing Video Reasoning in MLLMs' research paper. It is designed to test the logical and temporal inference capabilities of Multimodal Large Language Models.

Use Cases

• Benchmark MLLM reasoning capabilities using the video-based evaluation prompts
• Implement reinforcement learning for video understanding using the Video-R1 methodology
• Analyze temporal and causal reasoning in multimodal models using the provided test cases

Strengths

• Contains evaluation data for the Video-R1 model as described in the associated research
• Integrated with the tulerfeng/Video-R1 GitHub repository for code and model access
• Focuses on Multimodal Large Language Model (MLLM) reasoning in video contexts