ncic-ict/Yolov3_tricks

YOLOV3's pytorch edition with many Tricks embedded

Adding my new modules on Darknet has a lot of trouble to do,then i implement a full-edition yolov3 with many tricks on Pytorch

Open source object detection dataset of pedestrain and vehicles:FLIR

---

Performance

	mAP	Car	Person	bike
Official Baseline	74.3%	85.6%	79.4%	58.0%
Ours

TODO

• Focal loss
• GIOU loss
• CIOU loss
• Smooth L1 loss
• Soft NMS
• And NMS
• Merge NMS
• PANet
• Mosaic
• CutOut
• CutMix
• RandomFlip
• Mish

---

Example Projects

• https://github.com/zylo117/Yet-Another-EfficientDet-Pytorch
• https://github.com/ayooshkathuria/YOLO_v3_tutorial_from_scratch