Metraipolis › archive › dc_26f692e2e779
LEAP: Learned Block-wise Evidence Retrieval for Long Audio-Video Perception — arXiv:2609.39938
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.39938
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-07T12:45:58.200Z
- SHA-256 of the archived text
3c8759752873b256882840eb6759eaa51362522b33463e97c416fb9bfc7774f7- Stable URL of this page
https://metraipolis.com/documento/dc_26f692e2e779
How to cite
LEAP: Learned Block-wise Evidence Retrieval for Long Audio-Video Perception — arXiv:2609.39938 — arXiv (en), https://arxiv.org/abs/2609.39938 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_26f692e2e779 (Metraipolis archive dc_26f692e2e779, sha256 3c8759752873b256882840eb6759eaa51362522b33463e97c416fb9bfc7774f7)
BibTeX
@misc{dc26f692e2e779,
title = {LEAP: Learned Block-wise Evidence Retrieval for Long Audio-Video Perception — arXiv:2609.39938},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.39938},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_26f692e2e779; sha256 3c8759752873b256882840eb6759eaa51362522b33463e97c416fb9bfc7774f7},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.39938. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Hour-scale audio-visual question answering is constrained by a context dilemma: dense whole-recording encoding rapidly exhausts context limits, whereas uniform temporal compression severely dilutes fine-grained acoustic and visual evidence. We introduce LEAP, a framework where the model retrieves its own evidence without placing the whole recording in one context. LEAP divides a recording into fixed-duration blocks, applying a lightweight localization pass to each block to score short candidate windows. The highest-ranked windows are pooled and re-encoded in a single bounded answer pass. Consequently, the answer input and peak context remain independent of the recording duration. By decoupling evidence localization from reasoning, our framework can localize candidate temporal windows over pre-computed transcripts without decoding media frames, while preserving fine-grained visual and non-speech evidence by routing the final answering pass over raw audio-visual streams. LEAP trains both stages: a localization LoRA improves the selected windows, and an answer LoRA improves the answers read from the same windows. The block grid natively supports causal queries, enabling LEAP to support streaming inference without streaming-specific training. Across several AVQA benchmarks, LEAP improves over the Qwen3-Omni-30B-A3B baseline by 4.5-16.8%, and transfers to a second omni-modal backbone, MiniCPM-o 4.5, surpassing its published results by 3.1-13.0%. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-10-01 in cs.AI). Categorie arXiv: Computation and Language (cs.CL) ; Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV). Pagina del lavoro: https://arxiv.org/abs/2609.39938 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito). Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina. Letto il 2026-10-07T12:14:14.184Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-10-01?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).