Rich Human Feedback for Text-to-Image Generation · HackerPapers discussionshttps://hackerpaper.kaipfister.de/p/2312.10240/Collected HN comments and submissions linking this paper.enMon, 02 Nov 2026 11:27:26 GMTRich Human Feedback for Text-to-Image Generation — HN submissionhttps://news.ycombinator.com/item?id=42656632urn:hackerpapers:submission:2312.10240:42656632Fri, 10 Jan 2025 15:47:29 GMT<p>Show HN: We collected detailed annotations for text-to-image generation</p><p>Recently, the most popular modality text-to-image annotations has been preference data, where annotators usually choose between two images two indicate their favorite. While this does work to fine-tune models, it lacks additional information about what might be wrong with the images. E.g., what part of the image is misaligned relative to the prompt. Google research propose a modality for more information rich annotations (https://arxiv.org/abs/2312.10240). Based on this, we produced this dataset of ~13k images. We collected in total ~1.5 million annotations from 150k annotators using our…</p><p>2 points; 0 discussion replies on HN. <a href="https://hackerpaper.kaipfister.de/p/2312.10240/">Paper and source history</a>. A submission is not an endorsement.</p>cs.CV