CpasNet on 227*227 data and 196 classes

[tethys0221]

I'm trying to train this with images that are 227*227 and have 196 types. I wrote my own load_data() but when training, error "Resource exhausted, OOM when allocating tensor with shape [128,3872,196,16,1]" occurs. (128 is batch size)

[AlLongley]

This implementation of CapsNet (If not every other as well) is quite heavy in terms of memory overhead for weights/optimizer gradients.

You're supplying a 128 x 3872 x 196 x16 element batch input, as float32 values would take about 6.2 GB of memory, considering optimizer overhead I could see that being some multiple of that.

Does it run with a batch size of 1? Try 10 and see what difference it makes in memory usage, and work out your highest usable batch size like that