AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

Weakly-Supervised Text-Driven Contrastive Learning for Facial Behavior Understanding

Xiang Zhang, Taoyue Wang, Xiaotian Li, Huiyuan Yang, Lijun Yin

ICCV 2023 · Label-efficient and robust learning › Recognition and long-tail · approach: Contrastive representation theory · 25 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 activity-text weakly-supervised pre-training on top of image-only pre-training with label names and descriptions (ablation)CLEF · BP4Daverage F1 (%): 64.8 -> 65.9Improved
add activity-text pre-training (ablation)CLEF · RAF-DBaccuracy (%): 89.44 -> 90.09Improved
add label names to CLIP image encoder (ablation)CLIP · BP4Daverage F1 (%): 63.4 -> 64.0Improved
add label descriptions in addition to label names (ablation)CLIP · RAF-DBaccuracy (%): 88.72 -> 89.70Improved
add activity-text pre-training without label text (ablation)CLEF · BP4Daverage F1 (%): 65.0 -> 64.2Hurt
add label names only after image pre-training (ablation)CLEF · RAF-DBaccuracy (%): 89.67 -> 88.57Hurt
use a shared text encoder for label names and descriptions instead of two (ablation)CLEF · BP4Daverage F1 (%): 64.5 -> 65.9Improved
apply CLEF pretraining and text-driven fine-tuning vs FaRL in zero-shotFaRL · RAF-DB zero-shotaccuracy (%): 16.21 -> 29.14Improved
CLEF versus DMUE on AffectNet-8CLEF · AffectNet-8accuracy (%): 62.84 -> 62.77No 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