Advice on the max_aggregation_fn

[zhouzuguang]

It takes me about 12mins to train the Car model for each epoch with GTX 1070 (8G) using the TF version(https://github.com/WeijingShi/Point-GNN), but it takes me about 20mins to train the model with your PyTorch version. I find there may be some advances in the max_aggregation_fn function in the file model.py.

In this way, I can train the model in about 12mins each epoch with my computer.