OneEmo: A Unified Multimodal Reasoning Model for Emotion Perception, Understanding, and Interaction
OneEmo is pitched as a single affective model that handles emotion perception, comprehension, and interaction together.
The paper introduces EmoWorld-130K, a dataset built with human-in-the-loop reasoning traces for affective tasks. The authors say supervised tuning across those tasks produced mutual gains, then added Emo-Chord, a reinforcement learning method for multi-task reward allocation. In their experiments, OneEmo beats similarly sized baselines on most benchmarks and remains competitive with larger commercial models. HF Daily Papers' note
The paper introduces EmoWorld-130K, a dataset built with human-in-the-loop reasoning traces for affective tasks. The authors say supervised tuning across those tasks produced mutual gains, then added Emo-Chord, a reinforcement learning method for multi-task reward allocation. In their experiments, OneEmo beats similarly sized baselines on most benchmarks and remains competitive with larger commercial models. HF Daily Papers' note
score 4