Worse results than the paper displayed
JunyuanDeng opened this issue · 4 comments
I'm running the kitti 00 sequence with python API, but I got the result below:
which is much worse than the paper displayed:
I got this result without editing the config file.
And I repeated it by changing the fill_holes to True, but the result did not change.
Why there are two different results?
For example, there are overlaps of cars and bumpy roads.
Hello @Rotatingpencil, that looks like pose estimation errors. Usually the so-called "ground truth" poses from KITTI are not enough. You can give it a try with these poses:
00.txt
Just override the 00.txt on your filesystem. Please let me know if this helps with the problem
Now the result seems perfect!
How can I get the correct poses that you sent me, the poses I used before were downloaded from the official website, but it's wrong....
You can use any SLAM system, for example SuMa ;)