search

lvsn / 6DOF_tracking_evaluation

Code visualize and evaluate the dataset from "A Framework for Evaluating 6-DOF Object Trackers".
http://vision.gel.ulaval.ca/~jflalonde/projects/6dofObjectTracking/index.html
Other
36 stars 6 forks source link

Considerable Labeling error in the demo data #14

Closed greatwallet closed 3 years ago

greatwallet commented 3 years ago

Hi there!

I tried to download the dataset from the link. However, the link to the full dataset seems to be invalid. Therefore, I downloaded the demo data, which contains only a sequence of interative hard sequence of object clock. I tried to visualize the ground truth annotation by "visualize_sequence.py" with model "geometry.ply", but in vain. The "geometry.ply" seems to be impossible to load due to wrong format issue. Then I substitute it with ply file transformed from "Clock.obj" by Meshlab. The visualization seems not good, it looks like below:

000009

There seems to be a considerable gap between the rendered mask and the object. Is it because the "geometry.ply" is not defined in the same coordination frame as "Clock.obj", or this is the best precision it could get?

And also, would you mind offering access to the full dataset?

Thank you very much!

greatwallet commented 3 years ago

Problem solved! It is proved that the Clock.obj is not aligned with geometry.ply.

BTW, I solved the problem of geometry.ply's loading problem, and it's because of WinZip's failure of extraction. Now I use Ubuntu commands and everything became okay.