Hao
three solutions: 1. find a machine with more memory 2. smaller batch size 3. reimplement the data loading part with tensorflow dataset API hope it helps.
this paper can answer your question https://arxiv.org/pdf/1705.03820.pdf
it is GAN collapse, the default dataset and training setting in this repo should be fine.
You need to understand GAN-CLS before reading this code
please report you TF and TL version
it seems the npz is not match with your network architecture.
I didn't provide pre-trained model for this repo
if you use my code to train the model, and use the same code to load npz, but it fails, I don't have any idea about it ... maybe you...
yes we did. to train on different dataset, just simply write your own `data_loader.py` here is an exmple of data loader for bird dataset. enjoy ``` if xxx ... elif...
this method is quite old, there should be some methods designed for sign language datasets.