AI Paper Atlas

Research map / Video / Video-language understanding

VidLaDA: Bidirectional Diffusion Large Language Models for Efficient Video Understanding

Zhihao He, Tieyuan Chen, Kangyu Wang, Ziran Qin, Yang Shao, Chaofan Gan, Shijie Li, Zuxuan Wu, Weiyao Lin

ICML 2026 · Video › Video-language understanding · approach: Video LLMs with fine-grained understanding · 0 citations (OpenAlex)

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 MARS-Cache to VidLaDA CoT decodingVidLaDA (CoT) · LongVideoBenchthroughput (tokens/s): 2.1 -> 25.2Improved
add MARS-Cache to VidLaDA CoT decodingVidLaDA (CoT) · LongVideoBenchaccuracy: 59.8 -> 59.7No clear effect
use MARS-Cache instead of Dual-CacheVidLaDA (CoT) · LongVideoBenchthroughput (tokens/s): 18.5 -> 25.2Improved
add MARS-Cache to LLaDA-V CoT decodingLLaDA-V (CoT) · LongVideoBenchthroughput (tokens/s): 4.5 -> 37.9Improved
add MARS-Cache to LLaDA-V CoT decodingLLaDA-V (CoT) · LongVideoBenchaccuracy: 58.1 -> 59.6Improved
add MARS-Cache to VidLaDA CoT decodingVidLaDA (CoT) · MLVU testaccuracy: 50.2 -> 50.7Improved
add MARS-Cache to VidLaDA CoT decodingVidLaDA (CoT) · EgoSchema subsetaccuracy: 67.4 -> 67.0Hurt
train on only 1min-3min videos instead of the baseline mix (ablation)VidLaDA Stage 2 · LongVideoBenchaccuracy: 62.0 -> 59.8Hurt
train on 2min-30min videos (ablation)VidLaDA Stage 2 · Video-MMEaccuracy: 62.1 -> 64.1Improved
add temporal scaling warm-up stage with 64 frames (Stage 2) (ablation)VidLaDA · LongVideoBenchaccuracy: 59.5 -> 62.0Improved
add long-form video expansion (Stage 3) (ablation)VidLaDA · Average over 8 video benchmarksaverage accuracy: 56.9 -> 57.9Improved
add short-clip temporal pre-training (Stage 1) to LLaDA-V (ablation)LLaDA-V · Average over 8 video benchmarksaverage accuracy: 51.2 -> 55.4Improved

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