ResearchBenchmarksMulti-view canonical image generation on Human4DiT 50 rigged human avatarsFollow23.775PSNRMagicMan19.87520.887521.922.9125Dec 3, 2024Evaluation ResultsMethodMethodLinksPSNRSSIMLPIPSNormal LossMagicManInput=uses generated c...Input=uses generated canonical pose image as input2024.1223.7750.9110.10770.134AniGS2024.1223.1250.9070.10230.101CHAMPFine-tuned=fine-tuned...Fine-tuned=fine-tuned on the multi-view synthetic dataset2024.1220.0250.8940.1458—