Data Augmentation while training?

[anuragranj]

Where is the data augmentation module located that is used while training? I can see src/caffe/data_transformer.cpp which probably defines all the transformation functions. I was hoping if someone could point me to the part of the code where these augmentations are being applied before training. In particular, I wanted to get into details of what sort of augmentation/preprocessing is used before feeding the data to the training pipeline. Thanks

[nikolausmayer]

Hi, look for files in "src/caffe/layers" whose names contain "augmentation". The code is split into modules that generate augmentation parameters (e.g. where to crop), modules that augment images, and modules that augment flow groundtruth.

[anuragranj]

Thanks