Tone

Thanks for the interest of my work. Currently, I can not share the conversion and deployment codes on NVIDIA TX2. Some guidance: Since the correlation is custom layer which is...

I do not use onnx as an intermediate step but define a TRT model directly.

Please see the [demo.py](https://github.com/ltkong218/FastFlowNet/blob/8012621729335ab52fb2bb37df6b5312412343fc/demo.py#L47). 20 is multiplied back here.

Do you use ground truth flow label of your real scene to train FastFlowNet in a supervised manner or in an unsupervised way? For optical flow visualization, the scale factor...

I think taking FlowNet2's prediction as ground truth will lead to error accumulation, I suggest you try to adopt RAFT's prediction as ground truth label. For training FastFlowNet, you should...

Maybe your CUDA and PyTorch version are relative high. CUDA 10.0 with PyTorch 1.2.0 are OK.

Please follow [IRR-PWC](https://github.com/visinf/irr) to train the model. You can directly replace IRR-PWC with FastFlowNet, since they have the same pyramid structure. We train FastFlowNet with the same datasets and augmentations...

You can ask here and discuss

According to my experience, provided correlation package only supports PyTorch 0.4.1. I install it under Ubuntu 16.04, and my gcc version is 5.4.0. I think you can try to search...

Please refer to [Pytorch Correlation module](https://github.com/ClementPinard/Pytorch-Correlation-extension). This module supports newer versions of PyTorch, such as 1.2 and so on.