Recent Text-to-Image (T2I) generation models such as Stable Diffusion and Imagen have made significant progress in generating high-resolution images based on text descriptions. However, many generated images still suffer from issues such as artifacts/implausibility, misalignment with text descriptions, and low aesthetic quality. Inspired by the success of Reinforcement Learning with Human Feedback (RLHF) for large language models, prior works collected human-provided scores as feedback on generated images and trained a reward model to improve the T2I generation. In this paper, we enrich the feedback signal by (i) marking image regions that are implausible or misaligned with the text, and (ii) annotating which words in the text prompt are misrepresented or missing on the image. We collect such rich human feedback on 18K generated images (RichHF-18K) and train a multimodal transformer to predict the rich feedback automatically. We show that the predicted rich human feedback can be leveraged to improve image generation, for example, by selecting high-quality training data to finetune and improve the generative models, or by creating masks with predicted heatmaps to inpaint the problematic regions. Notably, the improvements generalize to models (Muse) beyond those used to generate the images on which human feedback data were collected (Stable Diffusion variants). The RichHF-18K data set will be released in our GitHub repository: https://github.com/google-research/google-research/tree/master/richhf_18k.
@article{arxiv.2312.10240,
title = {Rich Human Feedback for Text-to-Image Generation},
author = {Youwei Liang and Junfeng He and Gang Li and Peizhao Li and Arseniy Klimovskiy and Nicholas Carolan and Jiao Sun and Jordi Pont-Tuset and Sarah Young and Feng Yang and Junjie Ke and Krishnamurthy Dj Dvijotham and Katie Collins and Yiwen Luo and Yang Li and Kai J Kohlhoff and Deepak Ramachandran and Vidhya Navalpakkam},
journal= {arXiv preprint arXiv:2312.10240},
year = {2024}
}