AI Paper Atlas

Research map / Video / Video-language understanding

AVoCaDO: An Audiovisual Video Captioner Driven by Temporal Orchestration

Xinlong Chen, Yue Ding, Weihong Lin, Jingyun Hua, Linli Yao, Yang Shi, Bozhou Li, Qiang Liu, Yuanxing Zhang, Pengfei Wan, Liang Wang

ICLR 2026 · Video › Video-language understanding · approach: Long video understanding

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add AVoCaDO SFT on 107K temporally aligned audiovisual captions (ablation)Qwen2.5-Omni-7B · video-SALMONN-2 testsetTotal error rate: 57.1 -> 41.4Improved
add dialogue-based reward R_D in GRPO (ablation)AVoCaDO-SFT · video-SALMONN-2 testsetDialogue F1: 74.4 -> 76.5Improved
add checklist-based reward R_C in GRPO (ablation)AVoCaDO GRPO with R_D · video-SALMONN-2 testsetTotal error rate: 41.3 -> 37.3Improved
add length-regularized reward R_L in GRPO (ablation)AVoCaDO GRPO with R_D and R_C · video-SALMONN-2 testsetRepetition collapse (%): 3.9 -> 0.4Improved
add length-regularized reward R_L in GRPO (ablation)AVoCaDO GRPO with R_D and R_C · video-SALMONN-2 testsetTotal error rate: 37.3 -> 37.3No clear effect
fine-tune SFT model on the same 2K samples used for GRPO instead of using GRPO (ablation)AVoCaDO-SFT · video-SALMONN-2 testsetTotal error rate: 41.4 -> 43.0Hurt
add dialogue reward R_D in GRPO (ablation)AVoCaDO-SFT · Daily-Omni by captionaccuracy: 48.1 -> 49.5Improved
generate captions jointly from audio and video instead of concatenating separate captionsGemini-2.5-Pro captions · Daily-Omni pilot, QA from captionaccuracy gain: +15.8% average, +27.8% on AV Event AlignmentImproved
build SFT captions with a two-stage (separate then fuse) procedure instead of direct joint captioning (ablation)Gemini-2.5-Pro SFT data generation · SFT data qualityinformation coverageImproved

Later papers that compared with this method

2 later accepted papers report AVoCaDO as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy