got an error

[nanciwan]

Thank you for sharing, but when I run your code, there is an error: ValueError: Dimensions must be equal, but are 84840 and 16 for 'capsule_3/conv2/add_2' (op: 'Add') with input shapes : [84840,16,48], [84840,16,16,16,48]. I changed the input data, but nothing else changed. Could you give me some suggestions?

[andyweizhao]

Please check the tf version. This repo was implemented in tf 1.4.1 and python 2.7

[nanciwan]

Please check the tf version. This repo was implemented in tf 1.4.1 and python 2.7

Thank you for your reply. I have solved this problem. But then I had another problem: I used two 2080Tis to run the program, but it was still out of memory. I tried to compress the capsule, but the effect was not ideal, and did not achieve the accuracy you achieved. So I want to ask you what kind of equipment you use to implement this project?

[andyweizhao]

This repo should not have the out-of-memory issue, because the number of labels and batch size are quite small, meaning that the routing is not expensive to run. I ran capsule-A with 1 GPU card (nvidia 1080) if I remember correctly.

[nanciwan]

This repo should not have the out-of-memory issue, because the number of labels and batch size are quite small, meaning that the routing is not expensive to run. I ran capsule-A with 1 GPU card (nvidia 1080) if I remember correctly.

Okay, I'll check it again in case I miss any details. Thanks very much.

[andyweizhao]

Sure thing. Good luck!