A caffe implementation of MobileNet-YOLO (YOLOv2 base) detection network, with pretrained weights on VOC0712 and mAP=0.709
| Network | mAP | Download | Download | NetScope |
|---|---|---|---|---|
| MobileNet-YOLO-Lite | 0.675 | train | deploy | graph |
| MobileNet-YOLO | 0.709 | train | deploy | graph |
Note : Training from linux version and test on windows version , the mAP of MobileNetYOLO-lite was 0.668
Compare with YOLOv2
| Network | mAP | Weight size | Inference time (GTX 1080) |
|---|---|---|---|
| MobileNet-YOLO-Lite | 0.675 | 16.8 mb | 10 ms |
| MobileNet-YOLO | 0.709 | 19.4 mb | 24 ms |
| Tiny-YOLO | 0.57 | 60.5 mb | N/A |
| YOLOv2 | 0.76 | 193 mb | N/A |
Note : the yolo_detection_output_layer not be optimization , and batch norm and scale layer can merge into conv layer
- caffe training
- add pre-trained model
- fix bugs
- windows support
- Visual Studio 2013 or 2015
- CMake 3.4 or higher (Visual Studio and Ninja generators are supported)
- Anaconda
The build step was the same as MobileNet-SSD-windows
> cd $caffe_root
> script/build_win.cmd
Download yolov2 coco weights
Save at $caffe_root/models/convert
cd $caffe_root/models/convert
python weights_to_prototxt.py
Note : you may need specify python caffe path or copy python lib. here
cd $caffe_root
examples\demo_darknet19.cmd
Download deploy model
Save at $caffe_root/models/yolov2
> cd $caffe_root/
> examples\demo_yolo_lite.cmd
If load success , you can see the image window like this
Download lmdb
Unzip into $caffe_root/
Please check the path exist "$caffe_root\examples\VOC0712\VOC0712_trainval_lmdb"
Download pre-trained weights , and save at $caffe_root\model\convert
> cd $caffe_root/
> examples\train_yolo_darknet.cmd
> cd $caffe_root/
> examples\train_yolo.cmd
- yolov3