DIME: Scaling Facial Representation Learning via Differential Masked Autoencoding

Benchmark Model Rank Results
face-alignment-on-wflwDIME-B–NME (inter-ocular): 3.82AUC@10 (inter-ocular): 62.94…
face-alignment-on-wflwDIME-L–NME (inter-ocular): 3.74AUC@10 (inter-ocular): 63.74…
face-alignment-on-wflwDIME-S–NME (inter-ocular): 3.93AUC@10 (inter-ocular): 61.28…
face-parsing-on-celebamask-hqDIME-B–Mean F1: 89.93
face-parsing-on-celebamask-hqDIME-L–Mean F1: 90.12
face-parsing-on-celebamask-hqDIME-S–Mean F1: 89.61
face-parsing-on-lapaDIME-B–Mean F1: 94.15
face-parsing-on-lapaDIME-L–Mean F1: 94.27
face-parsing-on-lapaDIME-S–Mean F1: 93.86