aliushn/mobilenetv2-yolov3

A darknet implementation of MobileNetV2-YOLOv3 detection network,5.0 BFLOPS!!!&MobileNetV2-YOLOv3-Lite:2.0Bflops, HUAWEI Mate30 33ms!!!

MobileNetv2-YOLOv3-SPP Darknet

A darknet implementation of MobileNetv2-YOLOv3-SPP detection network

Network	VOC mAP(0.5)	Resolution	Inference Time(GTX2080ti)	FLOPS	Weight size
MobileNetV2-YOLOv3-SPP	71.7	416	5ms	5.5BFlops	14.2

*emmmm...这个懒得训练，mAP就凑合这样吧

Mobile inference frameworks benchmark (4*ARM_CPU)

Network	VOC mAP(0.5)	COCO mAP(0.5)	Resolution	Inference time (NCNN/Kirin 990)	Inference time (NNN arm82/Kirin 990)	FLOPS	Weight size
MobileNetV2-YOLOv3-Lite	72.61	35.2	320	33 ms	18 ms	2.1BFlops	9.8MB
MobileNetV2-YOLO-Tiny	61.17	30.4	304	26 ms	11 ms	1.5Flops	3.9MB
yolov3-tiny-prn	&	33.1	416	& ms	& ms	3.5BFlops	18.8MB

• Darknet Train Configuration: CUDA-version: 10010 (10020), cuDNN: 7.6.4,OpenCV version: 4 GPU:RTX2080ti
• Darknet Packet convolution is not well supported on some GPUs such as gtx1080ti, and the MobileNetV2-YOLOv3-SPP inference time is 100ms
• Support mobile inference frameworks such as NCNN&MNN

NCNN conversion tutorial

• benchmark:https://github.com/Tencent/ncnn/tree/master/benchmark
• darknet2ncnn: https://github.com/Tencent/ncnn/tree/master/tools/darknet

MNN conversion tutorial

• 待完成

Reference&Framework instructions

https://github.com/AlexeyAB/darknet