Awesome
nndet2
Object Detection models based on detectron2.
Backbones
Model | Backbone | Lr schd | Mem (GB) | mask AP | box AP | Config | Download |
---|
Mask RCNN | ConvNeXt_Tiny | 1x | | 40.5 | 43.8 | config | model |
RetinaNet | ConvNeXt_Tiny | 1x | | | 43.8 | config | model |
Model | Backbone | Lr schd | Mem (GB) | mask AP | box AP | Config | Download |
---|
RetinaNet+ | ResNet50* | 1x | | | 41.9 | config | model |
+
: we use GN + GIoU + multi-scale training
*
: we use new TorchVision SOTA ResNet models
Model | Backbone | Lr schd | Mem (GB) | mask AP | box AP | Config | Download |
---|
RetinaNet | DaViT-Tiny | 1x | | | 44.8 | config | model |
Model | Backbone | Lr schd | Mem (GB) | mask AP | box AP | Config | Download |
---|
RetinaNet | SwinV2-Tiny | 1x | | | 44.8 | config | model |