MinfengZhu / DM-GAN

MIT License
187 stars 64 forks source link

IS value is Nan #31

Open 405530042 opened 2 years ago

405530042 commented 2 years ago

I tried to run the IS code in tf19.12 and got some problem Here is my setting in inception_score_birds: tf.flags.DEFINE_string('checkpoint_dir','./model.ckpt', """Path where to read model checkpoints.""") tf.flags.DEFINE_string('image_folder', './bird_DMGAN/valid/single', """Path where to load the images """) tf.flags.DEFINE_integer('num_classes', 50, # 20 for flowers """Number of classes """) tf.flags.DEFINE_integer('splits', 10, """Number of splits """) tf.flags.DEFINE_integer('batch_size', 3, "batch size") and i turn the Line 79 img = scipy.misc.imresize(img, (299, 299, 3), interp='bilinear') into img = np.array(Image.fromarray(img).resize((299,299)))

cause misc.imresize does not work ,i have to change it.

i can surely catch images from the images_folder, but when I print the kl value ,some values in matrix will be Nan and the final values( mean and std) will be Nan in the end.Has anyone meet this problem and solved it? btw,when i ran the IS code, it always takes me half an hour or more,does anyone know how to improve it?