Are you sure .train() puts batch norm in eval mode? I'd assume that means train...

https://github.com/awslabs/aws-cv-task2vec/blob/c5795e55ba773f9845498091a90eee2fcba5da31/task2vec.py#L71

```python
# Fix batch norm running statistics (i.e., put batch_norm layers in eval mode)
self.model.train()
```

is this truly correct?

.train() usually puts layers in training mode. So for batch norm what it means is that it start collecting running statistics but uses mini-batch stats if I remember correctly, while in .eval() it uses the saved running stats. Right?