Query2Label: A Simple Transformer Way to Multi-Label Classification

Benchmark Model Rank Results
multi-label-classification-on-ms-cocoQ2L-CvT(ImageNet-21K pretraining, resolution 384)#2mAP: 91.3
multi-label-classification-on-ms-cocoQ2L-SwinL(ImageNet-21K pretraining, resolution 384)#5mAP: 90.5
multi-label-classification-on-ms-cocoQ2L-TResL(ImageNet-21K pretraining, resolution 640)#6mAP: 90.3
multi-label-classification-on-ms-cocoQ2L-R101(resolution 448)#22mAP: 84.9
multi-label-classification-on-nus-wideQ2L-CvT(resolution 384, ImageNet-21K pretrained)#1MAP: 70.1
multi-label-classification-on-nus-wideQ2L-TResL(resoluition 448)#3MAP: 66.3
multi-label-classification-on-nus-wideQ2L-R101(resolution 448)#5MAP: 65.0
multi-label-classification-on-pascal-voc-2007Q2L-CvT(ImageNet-21K pretrained, resolution 384)#1mAP: 97.3
multi-label-classification-on-pascal-voc-2007Q2L-TResL(ImageNet-21K pretrained, resolution 448)#2mAP: 96.9
multi-label-classification-on-pascal-voc-2007Q2L-TResL(resolution 448)#6mAP: 96.1