hwang1996 / ManiCLIP

ManiCLIP: Multi-Attribute Face Manipulation from Text
18 stars 1 forks source link

test_latents_seed100.pt #1

Closed MengZhen-Chi closed 4 months ago

MengZhen-Chi commented 1 year ago

could you tell me how to get the "test_latents_seed100.pt" under the data direction, through e4e encoder?

douxWh commented 4 months ago

could you tell me how to get the "test_latents_seed100.pt" under the data direction, through e4e encoder? Excuse me, have you got the answer?

MengZhen-Chi commented 4 months ago

could you tell me how to get the "test_latents_seed100.pt" under the data direction, through e4e encoder? Excuse me, have you got the answer?

nothing😂

hwang1996 commented 4 months ago

We did not use the inverted real images for evaluation.

The latents are randomly initialized nosie. They are used as the GAN generator's input to produce the original images. We set the ramdom seed as 100. Then we used the first sentences of the last 5,000 text files from Multi-Modal-CelebA-HQ dataset as the guidance to edit the original images for evaluation.

Thanks.

yuelang222 commented 1 week ago

Thanks to open source, how it works on the celebA-HQ 1024 dataset