search

tancik / StegaStamp

Invisible Hyperlinks in Physical Photographs
http://www.matthewtancik.com/stegastamp
MIT License
675 stars 191 forks source link

can this model handle random size image input #33

Closed Hammer-888 closed 3 years ago

Hammer-888 commented 3 years ago

as title

tancik commented 3 years ago

The code is currently set up for 400x400px images. You would need to make modifications and retrain if you wanted to encode a different resolution.

stephenivy07 commented 3 years ago

The code is currently set up for 400x400px images. You would need to make modifications and retrain if you wanted to encode a different resolution.

Hi, I modified the code and wanted to retrain it on 256x256px images, but I met some problems. Could you please give me some guidance on how to train on images that are not 400x400px? Thank you so much! 捕获

fujiaixin09 commented 3 years ago

The code is currently set up for 400x400px images. You would need to make modifications and retrain if you wanted to encode a different resolution.

Hi, I modified the code and wanted to retrain it on 256x256px images, but I met some problems. Could you please give me some guidance on how to train on images that are not 400x400px? Thank you so much! 捕获

Because in coding, the author uses the u-net structure, and the left downsampling part is connected with the right upsampling part, so the dimension you entered is not correct.

April4lu commented 3 years ago

The code is currently set up for 400x400px images. You would need to make modifications and retrain if you wanted to encode a different resolution.

Hi, I modified the code and wanted to retrain it on 256x256px images, but I met some problems. Could you please give me some guidance on how to train on images that are not 400x400px? Thank you so much! 捕获

I met the same error, did you solve this issue? I'll appreciate for your help!