AI Paper Atlas

Research map / Vision-language and multimodal / Cross-modal learning

AutoAD II: The Sequel - Who, When, and What in Movie Audio Description

Tengda Han, Max Bain, Arsha Nagrani, Gul Varol, Weidi Xie, Andrew Zisserman

ICCV 2023 · Vision-language and multimodal › Cross-modal learning · approach: Audio-visual speech separation · 42 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
use cross-attention conditioning instead of prompt-based visual conditioning (ablation)AD generation LM · MAD-eval-namedCIDEr: 6.7 -> 10.0Improved
add recognized character names to prompt (ablation)X-Attn AD model · MAD-eval-namedCIDEr: 10.0 -> 14.2Improved
add actor names to character prompt (ablation)X-Attn AD model · MAD-eval-namedCIDEr: 14.2 -> 15.0No clear effect
add character exemplar image features (ablation)X-Attn AD model · MAD-eval-namedCIDEr: 15.0 -> 18.3Improved
use recognized active characters instead of full cast list (ablation)X-Attn AD model · MAD-eval-namedCIDEr: 14.9 -> 18.3Improved
add AudioVault and WebVid partial-data pretraining (ablation)Ours · MAD-eval localCIDEr: 18.3 -> 19.2Improved
add previous AD as recurrent context (ablation)Ours with pretraining · MAD-evalCIDEr: 19.2 -> 19.5Improved
use transformer decoder instead of thresholding cosine similarity for character recognition (ablation)Character recognition module · 4 MAD-eval moviesAverage Precision: 0.55 -> 0.87Improved
train recognition on MovieNet face labels instead of NER-mined labels (ablation)Character recognition module · 4 MAD-eval moviesAverage Precision: 0.74 -> 0.85Improved
add linear projection layer to character recognition (ablation)Character recognition module · 4 MAD-eval moviesAverage Precision: 0.85 -> 0.87Improved
add visual features to duration for AD temporal proposal (ablation)TFM temporal proposal · MAD-t-evalAverage Precision: 0.53 -> 0.61Improved
add movie subtitles as model input (ablation)AD generation · MAD-evalCIDErNo clear effect

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