**This is an experimental project , try to use different backbone models with yolov3 detector
Train on VOC2007 and VOC2012 trainval and test on VOC2007
| network | mAP | resolution | macc | param |
|---|---|---|---|---|
| PVA-YOLOv3 | 0.703 | 416 | 2.55G | 4.72M |
| Pelee-YOLOv3 | 0.703 | 416 | 4.25G | 3.85M |
| DenseNet_121-YOLOv3 | 416 | 9.86G | 7.07M | |
| Resnet-18 | 0.714 | 416 | 6.33G | 11.24M |
- All of the models trained from imagenet pre-train weights , you can easlity to find download link from author's page
- These models are not optimized , but you can modify architecture and try to find best accuracy and inference speed.