AI Paper Atlas

Research map / Video / Video-language understanding

Thinking with Drafts: Speculative Temporal Reasoning for Efficient Long Video Understanding

Pengfei Hu, Meng Cao, Yingyao Wang, Yi Wang, Jiahua Dong, Jun Song, Yu Cheng, Bo Zheng, Xiaodan Liang

CVPR 2026 · Video › Video-language understanding · approach: Long video understanding · 3 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
replace uniform 16-frame sampling with target+draft frame selection (ablation)SpecTemp · Video-Holmesaccuracy: 39.5 -> 47.0Improved
add target-predicted temporal region before uniform sampling (ablation)Uniform sampling · Video-Holmesaccuracy: 39.5 -> 43.2Improved
use CLIP similarity for frame selection after region prediction (ablation)Target + Uniform · Video-Holmesaccuracy: 43.2 -> 45.1Improved
replace CLIP with a trained 3B draft model for frame selection (ablation)Target + CLIP · Video-MME (wo sub)accuracy: 60.5 -> 62.4Improved
use 7B-only for both roles instead of 7B target + 3B draft (ablation)SpecTemp · LongVideoBenchaccuracy and latency: 57.5 -> 54.1 acc; 2.3s -> 3.1sHurt
use 3B-only for both roles instead of 7B target + 3B draft (ablation)SpecTemp · LongVideoBenchaccuracy: 57.5 -> 40.3Hurt
add SFT to target model (ablation)Qwen2.5-VL-7B target · LongVideoBenchaccuracy: 54.5 -> 55.3Improved
add RL to target model after SFT (ablation)Target model · LongVideoBenchaccuracy: 55.3 -> 56.3Improved
add draft model SFT (ablation)Target SFT · LongVideoBenchformat accuracy: 95.2 -> 98.7Improved
add RL to draft model (full SFT+RL for both) (ablation)SpecTemp · LongVideoBenchaccuracy: 57.1 -> 57.5Improved
use SpecTemp at about 14 frames instead of Qwen2.5-VL-7B 16 framesQwen2.5-VL-7B · Video-Holmesaccuracy: 35.0 -> 47.0Improved
use SpecTemp instead of VideoChat-R1.5 at 64-frame scaleVideoChat-R1.5 · long-form latencylatency (s): 11.5 -> 8.9Improved

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