codeformer

codeformer : Towards Robust Blind Face Restoration with Codebook Lookup Transformer

Input

(Image from https://github.com/sanghyun-son/EDSR-PyTorch/blob/master/test/0853x4.png)

Ailia input shape : (1, 3, 512, 512)

Output

Ailia output shape : (1, 3, 512 , 512)

Usage

Automatically downloads the onnx and prototxt files when running. It is necessary to be connected to the Internet while downloading.

For the sample image with twice the resolution,

$ python3 codeformer.py

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

$ python3 codeformer.py --input IMAGE_PATH --savepath SAVE_IMAGE_PATH

If you want to specify the quality, put the scale after the --fidelity_weight option.

$ python3 codeformer.py --w 0.5

If you want to specify the cropped and aligned faces, put the scale after the --has_aligned option.

$ python3 codeformer.py --has_aligned

You can select a model from retinaface_resnet50 | retinaface_mobile0.25 by adding --arch

$ python3 codeformer.py --video VIDEO_PATH

Reference

Towards Robust Blind Face Restoration with Codebook Lookup Transformer

Framework

Pytorch 1.10.0

Model Format

ONNX opset = 10

Netron

codeformer.onnx.prototxt

face_parse.onnx.prototxt

retinaface_mobile0.25.onnx.prototxt

retinaface_resnet50.onnx.prototxt