Closed Armin234 closed 2 years ago
Thanks for the query! We have an instance of flow estimator network architecture implementation in our video super-resolution example (TecoGAN). https://github.com/sony/nnabla-examples/blob/master/video-superresolution/tecogan/models.py#L39
# Example usage
frames = F.concatenate(previous_frame, current_frame) # (N, H, W, C * 2)
flow = flow_estimator(frames) # (N, H, W, 2) 2 is (dx, dy)
(Note that the channel dimension comes last in this implementation.)
You may get the pre-trained flow estimator from the pretrained tecogan model, or by training the model from scratch using opticalflow dataset.
Sorry that unfortunately we don't have flownet implementation in our repository.
thank you, this should help me a lot
Hello, is there an example of FlowNet (Correlation) implementation with nnabla ? Or another simple example for optical flow estimation ?
Regards Armin