AI Paper Atlas

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

Label-anticipated Event Disentanglement for Audio-Visual Video Parsing

Jinxing Zhou, Dan Guo, Yuxin Mao, Yiran Zhong, Xiaojun Chang, Meng Wang

ECCV 2024 · Vision-language and multimodal › Cross-modal learning · approach: Self-supervised audio-visual representation learning · 20 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 MMIL decoder with LEAP label-projection decoderMM-Pyr · LLP test, event-level Type@AVF1: 57.1 -> 59.5Improved
replace MMIL decoder with LEAP label-projection decoderHAN · LLP test, event-level Type@AVF1: 56.4 -> 57.7Improved
add audio-visual semantic similarity loss L_avss (ablation)LEAP with MM-Pyr · LLP test, event-level Event@AVF1: 56.6 -> 57.4Improved
increase LEAP blocks from 1 to 2 (ablation)LEAP with MM-Pyr · LLP testaverage of ten metrics: 61.0 -> 61.3Improved
increase LEAP blocks from 2 to 4 (ablation)LEAP with MM-Pyr · LLP testaverage of ten metrics: 61.3 -> 61.4No clear effect
swap Glove label embeddings for Bert or CLIP (ablation)LEAP with MM-Pyr · LLP testaverage of ten metrics: 61.3 -> 60.9 (Bert), 61.1 (CLIP)No clear effect
use last block output rather than first or average (ablation)LEAP with MM-Pyr · LLP testaverage of ten metrics: 60.7 (first), 60.8 (average) -> 61.3 (last)Improved
set lambda to 1.0 instead of 0.5 or 2.0 (ablation)LEAP with MM-Pyr · LLP testaverage of ten metrics: 61.8 (0.5), 61.5 (2.0) -> 62.1Improved
use LEAP in place of the default decoder on overlapping eventsMM-Pyr · LLP overlapping subset, event-levelaverage F1: 48.1 -> 49.8Improved
replace default decoder with LEAP on audio-visual event localizationCMBS · AVE datasetaccuracy: 74.5 -> 77.9Improved

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