CaraJ/ORM-T2I-R1

This is the Output Reward Model (ORM) used in the paper T2I-R1: Reinforcing Image Generation with Collaborative Semantic-level and Token-level CoT.

T2I-R1 is a novel reasoning-enhanced text-to-image generation model powered by Reinforcement Learning (RL) with a bi-level Chain-of-Thought (CoT) reasoning process. This ORM is crucial for evaluating image generation by leveraging two levels of CoT:

1. 1.Semantic-level CoT: for high-level planning of the prompt.
2. 2.Token-level CoT: for low-level pixel processing during patch-by-patch generation.

The paper introduces BiCoT-GRPO with an ensemble of generation rewards, which seamlessly optimizes both generation CoTs within the same training step. By applying these reasoning strategies to the baseline model, Janus-Pro, T2I-R1 achieves superior performance with a 13% improvement on T2I-CompBench and 19% improvement on the WISE benchmark, even surpassing the state-of-the-art model FLUX.1.

This model is fine-tuned from lmms-lab/llava-onevision-qwen2-7b-ov.

For more details, please refer to the official paper and the GitHub repository.