search

amoodie / StratGAN

Coupling a generative adversarial network with image quilting to produce basin-scale realizations of fluvial stratigraphy.
MIT License
10 stars 2 forks source link

handle incomplete batch #2

Open amoodie opened 6 years ago

amoodie commented 6 years ago

need to come up with a way to handle an incomplete batch. This could be done by dynamically padding the Tensor with zeros to fill out the expected shape? Currently an error is thrown if drop_remainder == False

amoodie commented 6 years ago

Despite a flexible batch_size for use with the sampler method, there is still an error when the batch is smaller than the full batch size.

The following error is thrown in the first reshape of the generator:

Input to reshape is a tensor with 180 values, but the requested shape has 212