Long-to-Short Video Evidence Reasoning for Grounded Question Answering

We present LOVER, a \underline{L}ong to sh\underline{O}rt \underline{V}ideo \underline{E}vidence \underline{R}einforced model for grounded question answering (GQA). LOVER highlights three innovations over existing reinforcement-learning (RL) based video reasoning models: (1) $\mathbf{\text{Long-to-short}}\mathbf{\text{Video}}\mathbf{\text{Evidence}}\mathbf{\text{Curriculum}}\mathbf{\text{Learning}}$, which organizes RL training according to evidence duration and progressively adapts the model from long-range grounding to short-term reasoning; (2) $\mathbf{\text{GQA}}\mathbf{\text{Rewards}}$, which underscore the benefit of IoP reward over IoU for evidence spotting rather than strict temporal span overlap; (3) $\mathbf{\text{Adaptive}}\mathbf{\text{Timestamp}}\mathbf{\text{Rendering}}$, which adaptively renders timestamps onto video frames using background-aware position and color selection to enhance temporal observability. The three designs are model-agnostic and reciprocal. They effectively improve QA, grounding, and grounded QA performance over different backbones. Notably, LOVER built on Time-R1 achieves new state-of-the-art (SOTA) results among open-source models on popular GQA benchmarks: NExT-GQA and ReXTime. Comprehensive ablation studies further validate the effectiveness of our three innovative components.