Giters

tteepe / GaitGraph

Official repository for "GaitGraph: Graph Convolutional Network for Skeleton-Based Gait Recognition" (ICIP'21)

Home Page:https://arxiv.org/abs/2101.11228

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Questions about the data.

zhangzhi-git opened this issue · comments

commented

Hello author, thank you for your work. I would like to know how all the video frames recorded in the casia-b_all_frames.csv file are obtained? Because I observed that it may not only save each frame from the video. For example, the video file 015-cl-02-108.avi has 135 frames, but only 65 frames are recorded in your csv file.

commented

Hi,
you are right, my extraction was missing quite some frames...
I rerun the extraction, added the extraction script and updated the instructions (737b03e).

I am also rerunning the preprocessing and experiments. I will upload the new files and close the issue once it is done.

Cheers,
Torben

commented

Hi,

the missing frames were just empty frames that were pre filtered. But the detection stage will sort those frames out anyway.
I rerun everything and got the nearly same results with the full extraction.

Hope that clears it up.

Torben

commented

Thank you for your outstanding contributions in this field! I also have some doubts about data processing: although objects exist in the box frame with yolo3, will the entry and exit of objects in the beginning and end of a video lead to frames of mutilated bodies, which have an impact on skeleton estimation, such as skeleton collapse and large changes in nodes?