Why do you flip the data in evaluation?

[vinits5]

Hey @ydwen, Thanks for such an amazing work of SimPLE: Pairwise Similarity Learning and sharing the code for further research. It will be great if you help me understand below function and it's use.

Below function is used in validation step as well as in the main_inference.py file. Can you please explain following points?

1. Why are you flipping the data about axis [3]? (Data shape is BxCxHxW. So, the data will be flipped about the last axis which is width)
2. What is the intention behind concatenating feats and feats_flip? This creates a significant gap in the process of training and testing.