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
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add MARS-Cache to VidLaDA CoT decoding | VidLaDA (CoT) · LongVideoBench | throughput (tokens/s): 2.1 -> 25.2 | Improved |
| add MARS-Cache to VidLaDA CoT decoding | VidLaDA (CoT) · LongVideoBench | accuracy: 59.8 -> 59.7 | No clear effect |
| use MARS-Cache instead of Dual-Cache | VidLaDA (CoT) · LongVideoBench | throughput (tokens/s): 18.5 -> 25.2 | Improved |
| add MARS-Cache to LLaDA-V CoT decoding | LLaDA-V (CoT) · LongVideoBench | throughput (tokens/s): 4.5 -> 37.9 | Improved |
| add MARS-Cache to LLaDA-V CoT decoding | LLaDA-V (CoT) · LongVideoBench | accuracy: 58.1 -> 59.6 | Improved |
| add MARS-Cache to VidLaDA CoT decoding | VidLaDA (CoT) · MLVU test | accuracy: 50.2 -> 50.7 | Improved |
| add MARS-Cache to VidLaDA CoT decoding | VidLaDA (CoT) · EgoSchema subset | accuracy: 67.4 -> 67.0 | Hurt |
| train on only 1min-3min videos instead of the baseline mix (ablation) | VidLaDA Stage 2 · LongVideoBench | accuracy: 62.0 -> 59.8 | Hurt |
| train on 2min-30min videos (ablation) | VidLaDA Stage 2 · Video-MME | accuracy: 62.1 -> 64.1 | Improved |
| add temporal scaling warm-up stage with 64 frames (Stage 2) (ablation) | VidLaDA · LongVideoBench | accuracy: 59.5 -> 62.0 | Improved |
| add long-form video expansion (Stage 3) (ablation) | VidLaDA · Average over 8 video benchmarks | average accuracy: 56.9 -> 57.9 | Improved |
| add short-clip temporal pre-training (Stage 1) to LLaDA-V (ablation) | LLaDA-V · Average over 8 video benchmarks | average accuracy: 51.2 -> 55.4 | Improved |
Closest papers by meaning
- VideoMLA: Low-Rank Latent KV Cache for Minute-Scale Autoregressive Video Diffusion (NeurIPS 2026 · similarity 0.90)
- Mitigating Mask Prior Drift and Positional Attention Collapse in Large Diffusion Vision-Language Models (ICML 2026 · similarity 0.89)
- Attend Locally, Remember Linearly: Linear Attention as Cross-Frame Memory for Autoregressive Video Diffusion (NeurIPS 2026 · similarity 0.88)
- LLaDA-V: Large Language Diffusion Models with Visual Instruction Tuning (CVPR 2026 · similarity 0.88)
- V-CAST: Video Curvature-Aware Spatio-Temporal Pruning for Efficient Video Large Language Models (NeurIPS 2026 · similarity 0.88)
- FastVID: Dynamic Density Pruning for Fast Video Large Language Models (NeurIPS 2025 · similarity 0.88)
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.
