Exploring Plain Vision Transformer Backbones for Object Detection

Benchmark Model Rank Results
cross-domain-few-shot-object-detection-on-diorViTDeT-FT#4mAP: 29.4
cross-domain-few-shot-object-detection-on-neu-detViTDeT-FT#4mAP: 15.8
cross-domain-few-shot-object-detection-on-uoddViTDeT-FT#6mAP: 15.8
instance-segmentation-on-lvis-v1-0-valViTDet-H#6mask AP: 48.1mask APr: 36.9
instance-segmentation-on-lvis-v1-0-valViTDet-L#7mask AP: 46.0mask APr: 34.3
cross-domain-few-shot-object-detection-on-artaxorViTDeT-FT#7mAP: 23.4
cross-domain-few-shot-object-detection-on-clipark1kViTDeT-FT#7mAP: 25.6
object-detection-on-lvis-v1-0-valViTDet-H#7box AP: 53.4
object-detection-on-lvis-v1-0-valViTDet-L#9box AP: 51.2
cross-domain-few-shot-object-detection-on-deepfishViTDeT-FT#8mAP: 6.5
instance-segmentation-on-coco-minivalViTDet, ViT-H Cascade (multiscale)#10mask AP: 53.1
instance-segmentation-on-coco-minivalViTDet, ViT-H Cascade#16mask AP: 52
object-detection-on-coco-oViTDet (ViT-H)#10Average mAP: 34.3
object-detection-on-coco-oViTDet (ViT-H)#43Effective Robustness: 7.89
object-detection-on-coco-val2017ViTDet, ViT-H Cascade (multiscale)#20box AP: 61.3
object-detection-on-coco-val2017ViTDet, ViT-H Cascade#26box AP: 60.4