R1-Omni
Full-modality emotion recognition model — RL + verifiable rewards, explainable reasoning
A full-modality emotion recognition model with RL and verifiable rewards, and an explainable reasoning process.
Open foundation models — the base layers everything builds on.
Key features
- First application of RLVR to an omni-multimodal LLM for emotion recognition
- Outputs reasoning in <think> tags plus final answer in <answer> tags
- Built on HumanOmni-0.5B; multiple open checkpoints on Hugging Face
- Better out-of-distribution generalization than SFT baselines
Good for
- Emotion recognition from video and audio
- Explainable multimodal affective computing research
Stack & tags
emotion-recognitionmultimodalrlvropen-weightsomnimodalexplainableaffective-computing