@scalinglaw.eth
CVPR 2024 Best Paper Award: Rich Human Feedback for Text-to-Image Generation
- First dataset with detailed human feedback on generated images.
- Rich Automatic Human Feedback (RAHF) Model (Multimodal Transformer model to predict rich human feedback)
So, it enables that:
- Finetuning generative models using predicted scores.
- Region inpainting using predicted implausibility heatmaps.
- Using aesthetic scores for classifier guidance in diffusion models.
https://arxiv.org/abs/2312.10240
0 reply
0 recast
0 reaction