cs-xiao / targan Goto Github PK

Hi, I see in your paper you showed how well this network performs with different metrics (FID, S-score etc.)
Is there the possibility you will also upload the part of the code where you have done the evaluation in this repository?
Thank you

Hi, I tried your network, but I have some problems in the loading of the dataset, for instance the testset.
I noticed that the testset, downloaded from the chaos website, has no ground truth, so there is no folder named "ground". Therefore when you try to load the ChaosDataset_Syn_Test it will say:
modal:t1,fold:test/t1, total size:0
Is this a problem with the structure of the folder, or I have to split the train set that contains also the folder ground manually and not use at all the test set provided by chaos?

Hello there,
I am referring to #3
I tried to use the model genesis approach, but I have some doubt about the correct implementation.
After the global avarege pooling in the last encoder layer, the tensor in output has a shape of (#batch_size, 512).
What's the continue to have the correct FID results ?
Do you perform some mean and that's the FID?
Am I missing something?