AI Paper Atlas

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

Cross-Modal Supervision for Learning Active Speaker Detection in Video

Punarjay Chakravarty, Tinne Tuytelaars

ECCV 2016 · Vision-language and multimodal › Cross-modal learning · approach: Self-supervised audio-visual representation learning · 60 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 temporal-continuity weighting of samples to VAD-supervised person-specific training (ablation)Person-specific active speaker classifier (VAD) · Masters (speaker-specific)mean average AUC: 0.65 -> 0.82Improved
train person-specific classifiers with directional audio instead of a generic classifierActive speaker classifier · Mastersaverage AUC: 0.69 -> 0.81Improved
use VAD with hard-max posterior and no temporal weighting for person-specific training instead of directional audioPerson-specific classifier · Masters (speaker-specific)mean average AUC: 0.81 -> 0.65Hurt
weak supervision by VAD instead of full directional audio for the generic classifierGeneric active speaker classifier · Mastersaverage AUC: 0.69 -> 0.71No clear effect
add online adaptation with temporally weighted samples (up to 10 s per person)Prior generic classifier · ColumbiaAUC: about 5-15% improvementImproved
add temporal smoothing of thresholded scores up to about 3 secondsOnline-learnt classifier · ColumbiaF-score: plateauing of results at around 3 secondsImproved

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