Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces ChronoStitch, a training-free method designed to effectively compose visual key-value (KV) memories for long-video question answering by addressing the challenges of temporal reasoning. By re-basing stored post-rotary keys onto a global multimodal RoPE coordinate system, ChronoStitch preserves the necessary temporal structure and selectively recomputes high-deviation visual tokens to enhance event-ordering accuracy. The method demonstrates a significant performance improvement over naive composition techniques, achieving 3.3x faster processing than full joint re-prefilling while maintaining accuracy on benchmarks like Qwen2.5-VL-3B and TempCompass.
ChronoStitch achieves 3.3x faster long-video reasoning while improving event-ordering accuracy by effectively composing visual memories without retraining.
Long-video question answering requires a model to preserve visual evidence over time without repeatedly reprocessing the same video. A practical approach is to store the vision-language model's internal key-value (KV) cache for each video chunk and retrieve that state at query time. However, independently cached video chunks do not compose correctly: every chunk is prefilled from local rotary position zero, so naive concatenation collides temporal phases and removes the global order required for questions about what happened first, how often events occurred, or what changed across the video. This paper presents ChronoStitch, a training-free method for composing independently stored visual KV memories. The method first re-bases stored post-rotary keys onto a global three-axis multimodal RoPE coordinate system that preserves time, height, and width structure. We show why a one-dimensional scalar re-indexing is geometrically inconsistent for visual tokens because it turns spatial order within a frame into false temporal displacement. We then address the residual content gap left by positional repair: later chunks were originally encoded without attending to earlier chunks. ChronoStitch therefore selectively recomputes a small fraction of high-deviation later-chunk visual tokens while allowing them to attend over the composed cache. On Qwen2.5-VL-3B and the temporal split of TempCompass, ChronoStitch outperforms naive composition and position-only variants, improving event-ordering accuracy while running 3.3x faster than full joint re-prefilling.