AI Paper Atlas

Research map / Video / Video generation

Shot-by-Shot: Film-Grammar-Aware Training-Free Audio Description Generation

Junyu Xie, Tengda Han, Max Bain, Arsha Nagrani, Eshika Khandelwal, Gül Varol, Weidi Xie, Andrew Zisserman

ICCV 2025 · Video › Video generation · approach: Compositional text-to-video diffusion · 6 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 shot-based temporal context (2 context shots) with dynamic sampling and shot numbers (ablation)Training-free AD framework · CMD-ADCIDEr: 22.4 -> 25.1Improved
add shot number referral instead of no label or red frame box (ablation)Training-free AD framework · TV-ADCIDEr: 27.8 -> 28.9Improved
use dynamic rather than uniform frame sampling (ablation)Training-free AD framework · CMD-ADCIDEr: 24.1 -> 25.1Improved
inject thread structure into Stage I prompt (ablation)Training-free AD framework · TV-AD subsetCIDEr: 28.8 -> 30.7Improved
inject thread structure into Stage I prompt (ablation)Training-free AD framework · CMD-AD subsetCIDEr: 29.9 -> 30.7Improved
choose Stage I factors by predicted shot scale rather than all factors (ablation)Training-free AD framework · CMD-ADCIDEr: 26.0 -> 26.3Improved
add facial expression factor to Stage I (AutoAD-Zero style) (ablation)Training-free AD framework · CMD-ADCRITIC: 47.4 -> 46.8No clear effect
use dense spatial DINOv2 features instead of CLS tokens for thread clustering (ablation)Thread structure prediction · Thread-SafeAP: 0.786 -> 0.902Improved
compare temporally nearest frames (Side) rather than all or middle frames (ablation)Thread structure prediction · Thread-SafeAP: 0.822 -> 0.902Improved
swap in GPT-4o for both stagesTraining-free AD framework · TV-ADCIDEr: 31.1 -> 34.2Improved
generate five candidate ADs jointly and select the best (ablation)Training-free AD framework · TV-ADCIDEr: 31.1 -> 51.3Improved
increase sampling randomness (higher top-p and temperature) for independent candidates (ablation)Assisted AD generation · CMD-ADCIDEr: 33.3 -> 37.0Improved

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