The Capsnet MNIST accuracy is only <10% when running on my own computer but >95% when running the same Notebook on Google Colab. According to one issue the reason was in Cuda version 10.0 which was suggested to be resolved in 10.1 but I upgraded to 10.1 and the issue is still there.
Are there any more ideas what could be causing this problem?
I am running Pyton 3.7 and Tensorflow 1.15 with GPU
Hi
The Capsnet MNIST accuracy is only <10% when running on my own computer but >95% when running the same Notebook on Google Colab. According to one issue the reason was in Cuda version 10.0 which was suggested to be resolved in 10.1 but I upgraded to 10.1 and the issue is still there.
Are there any more ideas what could be causing this problem?
I am running Pyton 3.7 and Tensorflow 1.15 with GPU
// Daniel