search

NVIDIA / flownet2-pytorch

Pytorch implementation of FlowNet 2.0: Evolution of Optical Flow Estimation with Deep Networks
Other
3.12k stars 739 forks source link

The optical flow generated by two frames of the same image is not 0. The visual image has a lot of noise #239

Open WormPartner opened 3 years ago

Lifedecoder commented 3 years ago

What do you mean of visual image? Visualized optical flow?

WormPartner commented 3 years ago

What do you mean of visual image? Visualized optical flow?

hello! yeah, when input is two same frames, in other words, there is no moving target ,. and the output (visualized optical flow img ) is not a white img, there is a lot noise.

i find that in the train dataset, this situation is not considered! do you know how to solve this problem? thanks!

Lifedecoder commented 3 years ago

What do you mean of visual image? Visualized optical flow?

hello! yeah, when input is two same frames, in other words, there is no moving target ,. and the output (visualized optical flow img ) is not a white img, there is a lot noise.

i find that in the train dataset, this situation is not considered! do you know how to solve this problem? thanks!

I've tried it, yeah, same frames led to vague profile, but i suppose it's rational for a deep learning model. If you use traditional model to calculate optical flow pixel by pixel, result maybe zero.

WormPartner commented 3 years ago

What do you mean of visual image? Visualized optical flow?

hello! yeah, when input is two same frames, in other words, there is no moving target ,. and the output (visualized optical flow img ) is not a white img, there is a lot noise. i find that in the train dataset, this situation is not considered! do you know how to solve this problem? thanks!

I've tried it, yeah, same frames led to vague profile, but i suppose it's rational for a deep learning model. If you use traditional model to calculate optical flow pixel by pixel, result maybe zero

ok .really thanks!