Data augmentation layers
abrosua opened this issue · 2 comments
I have some question regarding the data augmentation methods and their parameters.
From my observation, the LiteFlowNet train prototxt (the data augmentation layer and their respective parameters, e.g. translate, rotate, zoom, color, contrast, brightness, etc) is very similar to the FlowNet train prototxt, instead of the FlowNet2.0 train prototxt.
Just to make sure, did you use the same data augmentation parameters as the one that written in the FlowNet paper? Because I find it quite hard to understand the caffe's augmentation layer cpp scripts, due to my lack of experiences with caffe.
Specifically we sample translation from the range [−20%,20%] of the image width for x and y; rotation from [−17, 17]; scaling from [0.9, 2.0]. The Gaussian noise has a sigma uniformly sampled from [0, 0.04]; contrast is sampled within [−0.8, 0.4]; multiplicative color changes to the RGB channels per image from [0.5, 2]; gamma values from [0.7,1.5] and additive brightness changes using Gaussian with a sigma of 0.2.
Thank you very much for your attention.
Regards
As shown in the prototxt file, data augmentation of LiteFlowNet and FlowNet2 are the same.
Sorry, I think I look at the wrong FlowNet2 train prototxt, the FlowNet2-S and FlowNet2-CSS train prototxt do share the similarity with the LiteFlowNet's. Thanks for the reply.