mobilenetv3

MobilenetV3

Input

Ailia input shape : (1,3,224,224)

Output

• large model

class_count=3
+ idx=0
  category=409[analog clock ]
  prob=10.01405143737793
+ idx=1
  category=892[wall clock ]
  prob=7.830683708190918
+ idx=2
  category=530[digital clock ]
  prob=3.3934028148651123

• small model

class_count=3
+ idx=0
  category=409[analog clock ]
  prob=19.85206413269043
+ idx=1
  category=892[wall clock ]
  prob=19.0461368560791
+ idx=2
  category=826[stopwatch, stop watch ]
  prob=12.139641761779785

usage

Automatically downloads the onnx and prototxt files on the first run. It is necessary to be connected to the Internet while downloading.

For the sample image,

$ python3 mobilenetv3.py

If you want to specify the input image, put the image path after the --input option.

$ python3 mobilenetv3.py --input IMAGE_PATH

By adding the --video option, you can input the video.
If you pass 0 as an argument to VIDEO_PATH, you can use the webcam input instead of the video file.

$ python3 mobilenetv3.py --video VIDEO_PATH

You can select a pretrained model by specifying -a large or -a small(default).

Reference

PyTorch Implemention of MobileNet V3

Model Format

ONNX opset = 10

Framework

Pytorch

Netron

mobilenetv3_small.onnx.prototxt mobilenetv3_large.onnx.prototxt