A caffe implementation of MobileNet-YOLO (YOLOv2 base) detection network, with pretrained weights on VOC0712 and mAP=0.709
| Network | mAP | Resolution | Download | NetScope |
|---|---|---|---|---|
| MobileNet-YOLO-Lite | 0.675 | 416 | deploy | graph |
| MobileNet-YOLOv3-Lite | 0.737 | 416 | deploy | graph |
| MobileNet-YOLOv3-Lite | 0.717 | 320 | deploy | graph |
Note :
- Training from linux version and test on windows version , the mAP of MobileNetYOLO-lite was 0.668
Compare with YOLOv2 , I can't find yolov3 score on voc2007 currently
| Network | mAP | Weight size | Inference time (GTX 1080) | Inference time (i5-4440) |
|---|---|---|---|---|
| MobileNet-YOLOv3-Lite | 0.717 | 20.3 mb | 6 ms (320x320) | 150 ms |
| MobileNet-YOLOv3-Lite | 0.737 | 20.3 mb | 11 ms (416x416) | 280 ms |
| Tiny-YOLO | 0.57 | 60.5 mb | N/A | N/A |
| YOLOv2 | 0.76 | 193 mb | N/A | N/A |
Note : the yolo_detection_output_layer not be optimization , and the deploy model was made by merge_bn.py
| mAP | Resolution | Download | NetScope |
|---|---|---|---|
| 53.9 | 416 | caffemodel | graph |
test on coco_minival_lmdb (IOU 0.5)
You can find non-depthwise convolution network here , Yolo-Model-Zoo
| network | mAP | resolution | macc | param |
|---|---|---|---|---|
| PVA-YOLOv3 | 0.703 | 416 | 2.55G | 4.72M |
| Pelee-YOLOv3 | 0.703 | 416 | 4.25G | 3.85M |
- 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
> 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"
> cd $caffe_root/
> examples\train_yolov3_lite.cmd
- COCO training and eval