mmfashion_tryon

MMFashion Virtual Try-on

Input

• Cloth image file
• Person image file
• Person-parse png image file created in palette mode that indexes 1, 2, 4, and 13 indicate the position of the head.
• Pose keypoints json file

(Image from VITON dataset https://drive.google.com/file/d/1MxCUvKxejnwWnoZ-KoCyMCXo3TLhRuTo/view)

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 mmfashion_tryon.py

If you want to specify a cloth-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 mmfashion_tryon.py --input CLOTH_IMAGE_PATH --savepath SAVE_IMAGE_PATH

If you want to specify a person image, put the image path after the -p option. Also, to specify a person-parse image, put the image path after the -pp option, and to specify the keypoint file, put the json file path after the -k option.

$ python3 mmfashion_tryon.py -p PERSON_IMAGE_PATH -pp PARSE_IMAGE_PATH -k JSON_FILE_PATH

If a person-parse image is unspecified, use the human-segmentation model to infer it.
And if a keypoint file is unspecified, use the pose-estimation model to infer it.

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

$ python3 mmfashion_tryon.py --video VIDEO_PATH --input CLOTH_IMAGE_PATH

Reference

• MMFashion

Framework

Pytorch

Model Format

ONNX opset=11

Netron

GMM_epoch_40.onnx.prototxt
TOM_epoch_40.onnx.prototxt