How to train with my own datasets? about co-mod-gan HOT 5 OPEN

The concatenation of DATA_DIR and DATASET should be the path to your tfrecords folder. For example, if the path of your tfrecords is ~/datasets/custom/custom.tfrecords, you can set --data-dir=~/datasets --dataset=custom

from co-mod-gan.

Thank you very much for your help. I'm a novice. Today I want to use ffhq data set for training, but I encountered the following problems. I would appreciate it if you could help me.

I have the original ffhq dataset (1024*1024，.png). Should I put these pictures in "/datasets"? And use "create_from_images.py" to generate into "/datasets/ffhq/ffhq-r10.tfrecords"? For example,I typed "python dataset_tools/create_from_images.py --tfrecord-dir=./datasets/ffhq --train-image-dir=./datasets --shuffle --compressed
" on the command line.

But then I used the files generated in this way for training, and the result failed. In addition, the file size of the ".tfrecords" I generated is 0KB, which I find very strange.

I'm sorry if there is any interruption.

from co-mod-gan.

You can directly download ffhq-r10.tfrecords without using create_from_images.py.

from co-mod-gan.

Thanks for your help. Unfortunately, I am in China and cannot download the dataset from Google Drive. So I can only make it myself.

from co-mod-gan.

OK I see... --compressed only works for JPEG images, so you can remove this option and try again.

from co-mod-gan.