Learning Emotion Representations from Verbal and Nonverbal Communication

TL;DR

EmotionCLIP is a novel pre-training framework that learns visual emotion representations from uncurated verbal and nonverbal communication data, addressing data scarcity and improving emotion recognition performance.

Contribution

This work introduces EmotionCLIP, the first pre-training paradigm for extracting emotion representations from communication data without curated labels, using subject-aware and sentiment-guided learning.

Findings

Outperforms state-of-the-art supervised methods in emotion recognition.

Rivals multimodal approaches across various benchmarks.

Addresses data scarcity in emotion understanding.

Abstract

Emotion understanding is an essential but highly challenging component of artificial general intelligence. The absence of extensively annotated datasets has significantly impeded advancements in this field. We present EmotionCLIP, the first pre-training paradigm to extract visual emotion representations from verbal and nonverbal communication using only uncurated data. Compared to numerical labels or descriptions used in previous methods, communication naturally contains emotion information. Furthermore, acquiring emotion representations from communication is more congruent with the human learning process. We guide EmotionCLIP to attend to nonverbal emotion cues through subject-aware context encoding and verbal emotion cues using sentiment-guided contrastive learning. Extensive experiments validate the effectiveness and transferability of EmotionCLIP. Using merely linear-probe evaluation protocol, EmotionCLIP outperforms the state-of-the-art supervised visual emotion recognition methods and rivals many multimodal approaches across various benchmarks. We anticipate that the advent of EmotionCLIP will address the prevailing issue of data scarcity in emotion understanding, thereby fostering progress in related domains. The code and pre-trained models are available at https://github.com/Xeaver/EmotionCLIP.