CapsNet implementations with PyTorch. Randomness in codes is removed.
Python 3.5
PyTorch 0.4.1
Sabour S, Frosst N, Hinton G E. Dynamic routing between capsules[C]//Advances in neural information processing systems. 2017: 3856-3866.
With the best hyperparameters on MNIST dataset in the paper, training will takes 5 hours with two GTX 1070.
| Best Test Accuracy | |
|---|---|
| Paper | 99.75% |
| This implementation with fine-tune | 99.74% |
| This implementation | 99.72% |
| Implementation by @gram-ai | 99.70% |
| Implementation by @XifengGuo | 99.66% |
| Implementation by @naturomics | 99.64% |
Hinton G E, Sabour S, Frosst N. Matrix capsules with EM routing[C]//ICLR 2018.
With the best hyperparameters on smallNORB dataset in the paper, training will takes 40 hours with eight Tesla P40.
| Best Test Accuracy | Training Speed | |
|---|---|---|
| Paper | 98.2% | - |
| This Implementation | 90.9% | < 1s/iter with 8 Tesla P40 |
| Implementation by @www0wwwjs1 | 91.8% | 25s/iter with 1 Tesla P40 |