deep-image-matting

Deep Image Matting

input

input image

input trimap

(from https://github.com/foamliu/Deep-Image-Matting-PyTorch/tree/master/data/alphamatting/input_lowres)

Ailia input shape: (1, 320, 320, 4) input range: (0,1) input color order : BGRA(keras), RGBA(torch)

output

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 deep-image-matting.py

If you want to specify the input image, put the image path after the --input and --trimap option.
You can use --savepath option to change the name of the output file to save.

$ python3 deep-image-matting.py --input IMAGE_PATH --trimap TRIMAP_PATH --savepath SAVE_IMAGE_PATH

You can use pytorch version with -f torch option. Currently the pytorch model requires ONNX Runtime. ailia SDK will support maxunpool operator.

$ python3 deep-image-matting.py -f pytorch --onnx

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. The trimap is generated by DeeplabV3.

$ python3 deep-image-matting.py --video VIDEO_PATH

Reference

Deep Image Matting

Deep-Image-Matting-PyTorch

Framework

Keras 2.3.1 or Pytorch 1.7.1

Model Format

ONNX opset = 10 (Keras) 11 (Pytorch)

Netron

deep-image-matting.onnx.prototxt deep-image-matting-pytorch.onnx.prototxt