Could you release the validation metrics of the flux version for us to reproduce and refer to #129
Description
Hello, may I ask if it is possible to disclose the test performance of the model's flux on the dataset? I tried to validate the flux version on the pair data of the VITON-HD dataset, setting the resolution to 512 high, 384 wide, and the repair to True during inference. Subsequently, the eval script was used for validation, and it was found that the metrics were very poor, with only:
{
"FID": 7.889195273779052,
"KID": 1.2820719119119075,
"SSIM": 0.8441665768623352,
"LPIPS": 0.09195264428853989
}
I don't know if my process was wrong. May I ask if you could release the validation metrics of the flux version for us to reproduce and refer to