search

minar09 / ACGPN

"Towards Photo-Realistic Virtual Try-On by Adaptively Generating↔Preserving Image Content"，CVPR 2020. (Modified from original with fixes for inference)
https://github.com/switchablenorms/DeepFashion_Try_On
77 stars 70 forks source link

How to generate test_label, test_edge, test_pose for custom image? #16

Open hrushikesh26 opened 3 years ago

hrushikesh26 commented 3 years ago

Hi, Thank you for this repo. This work pretty nicely. I am trying to test this on different images of my own So, how can I generate test_label, test_pose and test_edge folders given human image in test_img folder and dress image in test_color folder.

minar09 commented 3 years ago

Hi @hrushikesh26 , I have used this repository only for testing the VITON dataset. I guess, the way the repository is made, it's a bit long procedure to apply for the custom images. Have you tried referring to the original repository (https://github.com/switchablenorms/DeepFashion_Try_On) or the original paper authors? It would be better if you can hear their opinion.

As they used the Viton dataset, so probably, test_pose is generated with the OpenPose (COCO-18) format, test_label from LIP_JPPNet, and test_edge is maybe generated with OpenCV binary thresholding method. However, I think, for the test_label in the ACGPN training/testing, they applied an extra label-mapping (as discussed here: https://github.com/switchablenorms/DeepFashion_Try_On/issues/15). These are just my guesses, so before trying, please refer to the original repository/authors.

Good luck! Thank you.