From RGB Generation to Dense Field Readout: Pixel-Space Dense Prediction with Text-to-Image Models

Benchmark Model Rank Results
monocular-depth-estimation-on-kitti-relative-log-alignedReChannel-9B#2AbsRel: 0.063delta1: 0.959
monocular-depth-estimation-on-scannet-relative-log-alignedReChannel-9B#2AbsRel: 0.047delta1: 0.976
human-pose-keypoint-estimation-on-coco-val2017ReChannel-9B#11AP: 79.2AP50: 92.7
human-pose-keypoint-estimation-on-coco-val2017ReChannel-4B#22AP: 78.0AP50: 91.4