Beyond Appearance: a Semantic Controllable Self-Supervised Learning Framework for Human-Centric Visual Tasks

Benchmark Model Rank Results
pedestrian-attribute-recognition-on-pa-100kSOLIDER#5Accuracy: 86.38
pedestrian-detection-on-citypersonsSOLIDER#6Reasonable MR^-2: 9.7Heavy MR^-2: 39.4
person-re-identification-on-market-1501SOLIDER#7Rank-1: 96.9mAP: 93.9
person-re-identification-on-market-1501SOLIDER (RK)#11Rank-1: 96.7mAP: 95.6
person-re-identification-on-msmt17SOLIDER (with re-ranking)#2mAP: 86.5Rank-1: 91.7
person-re-identification-on-msmt17SOLIDER (without re-ranking)#5mAP: 77.1Rank-1: 90.7
person-re-identification-on-occluded-dukemtmcSOLIDER#6Rank-1: 71.2mAP: 61.9
person-search-on-cuhk-sysuSOLIDER#6MAP: 95.5Top-1: 95.8
person-search-on-prwSOLIDER#2mAP: 59.8Top-1: 86.7
pose-estimation-on-cocoSOLIDER (swin-B)#8AP: 76.6AR: 81.5
semantic-segmentation-on-lip-valSOLIDER#4mIoU: 60.50%