Currently , this model can not run on TensorRT , do not use
network | mAP (IOU 0.5) | resolution | netscope | Inference time |
---|---|---|---|---|
MobileNet-YOLOv3-Lite | 34.2 | 320 | graph | 3.20 ms |
MobileNet-YOLOv3-Lite | 37.5 | 416 | graph | 4.78 ms |
- MobileNet-YOLOv3-Lite pre-trained weights use caffemodel
- Remove group convolution when upsample
- Remove maxpooling
- Add 1x1 convolution