tfzhou / MATNet

Motion-Attentive Transition for Zero-Shot Video Object Segmentation (AAAI2020&TIP2021)

Home Page:https://www.researchgate.net/publication/343623463_MATNet_Motion-Attentive_Transition_Network_for_Zero-Shot_Video_Object_Segmentation

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

About Optical Flow

BruceWang1996 opened this issue · comments

First frame doesn't have optical flow picture, how to deal with this?

The folder 3rdparty/pytorch-pwc is empty, how did you generate optical flow?

@hailyxu Please refer to prepare-optical-flow in readme for how to run optical flow for your data.

@BruceWang1996 Actually, we don't have optical flow for the last frame. In our experiments, we simply ignore the last frame, which also follows the standard protocol in DAVIS.

@hailyxu Please refer to prepare-optical-flow in readme for how to run optical flow for your data.

Thanks for your reply. I have checked the folder 3rdparty/pytorch-pwc, but it's empty, there's no run_davis2017.py in it. I have also go to the original pytorch-pwc repo, but apparently, the code could only produce the optical flow of two frames once.
I would apprecirate it if you could offer some help about optical flow, it's really time comsuming if I go to find other models that produce optical flow results.

@hailyxu I have updated the optical flow helper scripts in the 3rdparty. Please update it or refer to the commit 7a996c1.

@hailyxu I have updated the optical flow helper scripts in the 3rdparty. Please update it or refer to the commit 7a996c1.

Thanks. Good day.