AI Paper Atlas

Research map / Video / Video-language understanding

FrameOracle: Learning What to See and How Much to See in Videos

Chaoyu Li, Tianzhi Li, Fei Tao, ZHENYU ZHAO, Ziqian Wu, Maozheng Zhao, Juntong Song, Cheng Niu, Pooyan Fazli

ICML 2026 · Video › Video-language understanding · approach: Long video 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 FrameOracle selection (16 -> 10.4 frames)LLaVA-Video-7B · 6 video benchmarks avgavg accuracy: 54.6 -> 54.7No clear effect
add FrameOracle selection from 64 candidates (13.9 frames)LLaVA-OneVision-7B · 6 video benchmarks avgavg accuracy: 49.8 -> 51.6Improved
select frames with FrameOracle instead of uniform 10 of 16 (ablation)LLaVA-OneVision-7B · avg accuracyaccuracy: 46.3 -> 50.3Improved
use FrameOracle selection on 16-frame inputLLaVA-OneVision-7B · NExTQA OE valaccuracy: 14.6 -> 16.1Improved
add FrameOracle (16 -> 10.4 frames)LLaVA-Video-7B · cost per samplelatency: 0.615 -> 0.363 sImproved
increase raw frame budget from 16 to 64LLaVA-Video-7B · avg accuracyavg accuracy: 54.6 -> 56.6Improved
add Stage 1 text-visual alignment alone (ablation)FrameOracle · 6 benchmarks avgavg accuracy: 50.5 -> 48.5Hurt
add Stage 2 frame ranking (ablation)FrameOracle · Qwen2.5-VL-3Bavg accuracy: 48.5 -> 50.8Improved
add Stage 3 K Head calibration (ablation)FrameOracle · Qwen2.5-VL-3Bavg accuracy: 46.9 -> 52.8Improved
add Stage 4 fine-tuning on FrameOracle-41K (ablation)FrameOracle · Qwen2.5-VL-3B avgavg accuracy: 52.7 -> 52.8Improved
train Stage 2 and 3 jointly instead of staged (ablation)FrameOracle · Rank head qualityKendall tau: 0.5367 -> 0.2313Hurt
skip intermediate stages (Stage 1+4 only, fixed K) (ablation)FrameOracle · Qwen2.5-VL-3B avgavg accuracy: 50.8 -> 49.6Hurt

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 0 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