search

dorarad / gansformer

Generative Adversarial Transformers
MIT License
1.33k stars 149 forks source link

`maps_in` in the graph? #21

Closed Susmit-A closed 2 years ago

Susmit-A commented 3 years ago

Hi,

I'm trying to extract some intermediate values from the model. The graph requires 3 inputs, as listed by Gs.list_layers() and Gs.input_names:

('latents_in', <tf.Tensor 'Gs/latents_in:0' shape=(?, 17, 32) dtype=float32>, [])
('labels_in', <tf.Tensor 'Gs/labels_in:0' shape=(?, 0) dtype=float32>, [])
('maps_in', <tf.Tensor 'Gs/maps_in:0' shape=(?, 16, 256, 256) dtype=float32>, [])

latents_in is the sampled z vector, and it appears labels_in is not required (the last dimension is 0). However, I'm not able to understand what maps_in is. How can I get maps_in?

dorarad commented 3 years ago

Hi, Thank you for the interest in the work! I have couple deadlines over the next days so will definitely try to get back to you by the end of the week!

dorarad commented 2 years ago

Hi, apologies for the large delay in my response and thanks so much for raising that point! maps_in refers to a unused remnant that stayed in the pickle and doesn't impact the model's behavior. I recommend loading the model using --recompile and then it should save a new snapshot that doesn't have the field. I will also update the checkpoints to not include that unnecessary field!

dorarad commented 2 years ago

Just thought to update I updated the pretrained networks files to match the source code so maps_in doesn't appear anymore!