Paper 2510.20579
Open-o3-Video: Grounded Video Reasoning with Explicit Spatio-Temporal Evidence
- Published
- Oct 2025
- Research lab
- Independent
- Citations
- 43
- GitHub
- 159 stars
01 In brief
Summary
Open-o3-Video is a non-agent framework that integrates explicit spatio-temporal evidence into video reasoning by highlighting key timestamps, objects, and bounding boxes.
The authors construct two datasets, STGR-CoT-30k and STGR-RL-36k, combining existing temporal and spatial grounding resources with 5.9k newly annotated spatio-temporal samples.
They adopt a two-stage training strategy: cold-start supervised fine-tuning followed by reinforcement learning using Group Sequence Policy Optimization (GSPO) with rewards for accuracy, thinking (temporal and spatial), and format.
Two novel mechanisms, adaptive temporal proximity and temporal gating, address reward sparsity and spatial collapse.
On the V-STAR benchmark, Open-o3-Video achieves state-of-the-art performance, improving mAM by 14.4% and mLGM by 24.2% over the Qwen2.5-VL baseline, and surpassing GPT-4o.
It also shows consistent gains on VideoMME, WorldSense, VideoMMMU, LongVideo-Reason-eval, and TVGBench.
The generated evidence supports confidence-aware test-time scaling, improving answer reliability.
02 From the paper
Abstract
Most video reasoning models only generate textual reasoning traces without indicating when and where key evidence appears. Recent models such as OpenAI-o3 have sparked wide interest in evidence-centered reasoning for images, yet extending this ability to videos is more challenging due to the need for joint temporal tracking and spatial localization across dynamic scenes. We introduce Open-o3-Video, a non-agent framework that integrates explicit spatio-temporal evidence into video reasoning by highlighting key timestamps, objects, and bounding boxes, making the reasoning process traceable and verifiable. To enable this capability, we first construct high-quality datasets STGR that provide unified spatio-temporal supervision, which is absent in existing resources. We further adopt a cold-start reinforcement learning strategy with specially designed rewards that jointly encourage answer accuracy, temporal alignment, and spatial precision. On the V-STAR benchmark, Open-o3-Video achieves state-of-the-art performance, improving mAM by 14.4% and mLGM by 24.2% over the Qwen2.5-VL baseline, and shows consistent gains across a range of video understanding benchmarks. Beyond accuracy, the grounded reasoning traces produced by Open-o3-Video support confidence-aware test-time scaling, improving answer reliability.