Henrique Morimitsu

These values are obtained after combining two factors: - The flow value predicted by PWCNet is 20 times smaller than the original value. - The stride of the layer where...

The stride of a layer is a scalar value which represents how many times smaller that layer is compared to the input image. For example, if the input image size...

I don't know if it helps you, since I reimplemented part of the code myself (although it should still use the same procedure as this original RAFT). After training RAFT...

I just used the same parameters used to train the larger model.

I have a personal project for optical flow which includes RAFT here: https://github.com/hmorimitsu/ptlflow. It includes some examples in Google Colab. I hope this can help you and others to better...