Add model card and metadata for R1-Omni-0.5B
#1
by
nielsr
HF staff
- opened
README.md
ADDED
|
@@ -0,0 +1,67 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: cc-by-nc-4.0
|
| 3 |
+
library_name: transformers
|
| 4 |
+
pipeline_tag: video-text-to-text
|
| 5 |
+
---
|
| 6 |
+
|
| 7 |
+
# R1-Omni: Explainable Omni-Multimodal Emotion Recognition with Reinforcement Learning
|
| 8 |
+
|
| 9 |
+
[](https://modelscope.cn/models/iic/R1-Omni-0.5B)
|
| 10 |
+
[](https://huggingface.co/StarJiaxing/R1-Omni-0.5B)
|
| 11 |
+
[](https://arxiv.org/abs/2503.05379)
|
| 12 |
+
|
| 13 |
+
This model utilizes Reinforcement Learning with Verifiable Reward (RLVR) to perform omni-multimodal emotion recognition. Built upon the HumanOmni-0.5B model, R1-Omni excels at understanding visual and audio cues for emotion identification, even in out-of-distribution scenarios.
|
| 14 |
+
|
| 15 |
+
## 📖 Introduction
|
| 16 |
+
**R1-Omni** is the first application of Reinforcement Learning with Verifiable Reward (RLVR) to an Omni-multimodal large language model. It focuses on emotion recognition, where visual and audio modalities play crucial roles. Key insights include:
|
| 17 |
+
|
| 18 |
+
1) **Enhanced Reasoning Capability**: R1-Omni demonstrates superior reasoning abilities, enabling a clearer understanding of how visual and audio information contribute to emotion recognition.
|
| 19 |
+
2) **Improved Understanding Capability**: Compared to SFT, RLVR significantly boosts performance on emotion recognition tasks.
|
| 20 |
+
3) **Stronger Generalization Capability**: RLVR models exhibit markedly better generalization capabilities, particularly excelling in out-of-distribution scenarios.
|
| 21 |
+
|
| 22 |
+
## 📦 Model Download
|
| 23 |
+
The model is based on the open-source HumanOmni-0.5B model. The following models are available: HumanOmni-0.5B, the cold-start model EMER-SFT, the model MAFW-DFEW-SFT fine-tuned directly on the MAFW and DFEW training sets, and the final model R1-Omni.
|
| 24 |
+
|
| 25 |
+
<div align="center">
|
| 26 |
+
|
| 27 |
+
| **Model** | **HuggingFace** | **ModelScope** |
|
| 28 |
+
|------------------------|---------------------------------------------------------------------------------|-------------------------------------------------------------------------|
|
| 29 |
+
| `HumanOmni-0.5B` | [](https://hf.co/StarJiaxing/HumanOmni-0.5B) | [](https://modelscope.cn/models/iic/HumanOmni-0.5B) |
|
| 30 |
+
| `EMER-SFT` | [](https://hf.co/StarJiaxing/EMER-SFT-0.5B) | [](https://modelscope.cn/models/iic/EMER-SFT-0.5B) |
|
| 31 |
+
| `MAFW-DFEW-SFT` | [](https://huggingface.co/StarJiaxing/MAFW-DFEW-0.5B) | [](https://modelscope.cn/models/iic/MAFW-DFEW-0.5B) |
|
| 32 |
+
| `R1-Omni` | [](https://huggingface.co/StarJiaxing/R1-Omni-0.5B) | [](https://modelscope.cn/models/iic/R1-Omni-0.5B) |
|
| 33 |
+
</div>
|
| 34 |
+
|
| 35 |
+
## 🏆 Performance
|
| 36 |
+
|
| 37 |
+
Below are the performance on emotion recognition datasets. We use symbols to indicate whether the data is **in-distribution (⬤)** or **out-of-distribution (△)**.
|
| 38 |
+
|
| 39 |
+
| Method | DFEW (WAR) ⬤ | DFEW (UAR) ⬤ | MAFW (WAR) ⬤ | MAFW (UAR) ⬤ | RAVDESS (WAR) △ | RAVDESS (UAR) △ |
|
| 40 |
+
|----------------------------------|---------------|---------------|---------------|---------------|------------------|------------------|
|
| 41 |
+
| HumanOmni-0.5B | 22.64 | 19.44 | 20.18 | 13.52 | 7.33 | 9.38 |
|
| 42 |
+
| EMER-SFT | 38.66 | 35.31 | 38.39 | 28.02 | 29.00 | 27.19 |
|
| 43 |
+
| MAFW-DFEW-SFT | 60.23 | 44.39 | 50.44 | 30.39 | 29.33 | 30.75 |
|
| 44 |
+
| R1-Omni | 65.83 | 56.27 | 57.68 | 40.04 | 43.00 | 44.69 |
|
| 45 |
+
|
| 46 |
+
|
| 47 |
+
|
| 48 |
+
### Legend
|
| 49 |
+
- **⬤**: Indicates **in-distribution data** (DFEW and MAFW).
|
| 50 |
+
- **△**: Indicates **out-of-distribution data** (RAVDESS).
|
| 51 |
+
|
| 52 |
+
## 🛠️ Environment Setup
|
| 53 |
+
Our code is built on the R1-V framework. To set up the environment, please follow the installation instructions in the [R1-V repository](https://github.com/Deep-Agent/R1-V/)
|
| 54 |
+
|
| 55 |
+
## 🔍 Inference
|
| 56 |
+
Our inference code is based on the implementation from **HumanOmni**.
|
| 57 |
+
|
| 58 |
+
## 📚 Citation
|
| 59 |
+
If you find our work helpful, feel free to cite us.
|
| 60 |
+
```
|
| 61 |
+
{zhao2025r1omniexplainableomnimultimodalemotion,
|
| 62 |
+
title={R1-Omni: Explainable Omni-Multimodal Emotion Recognition with Reinforcement Learning},
|
| 63 |
+
author={Jiaxing Zhao and Xihan Wei and Liefeng Bo},
|
| 64 |
+
journal={arXiv preprint arXiv:2503.05379},
|
| 65 |
+
year={2025}
|
| 66 |
+
}
|
| 67 |
+
```
|