A distributed implementation of CapsNet for training and inference. Some optimization of network structure is also added for acceleration.
Status:
- The implementation of distributed training is finished.
- Finish the speed test on single GPU.
Daily task
- Validation performance with multi-gpus
- Implement the test and inference part
- Python
- NumPy
- Tensorflow 1.2.0+
Speed test report With single GPU GTX 1080 and CPU i7-5820K CPU @ 3.30GHz. 1 epoch for training on MNIST costs 157.4s, approximately 0.34s/iteration. 100 times inferences costs 4.2s, approximately 0.04s for inference.