bad result on kitti about dpvo HOT 7 OPEN

Yeah that doesn’t look good, thanks for bringing this up. It looks like there is some bias in the driving scenarios causing the scale to constantly expand.

I will look into this and reply to this thread in a few days. There might be a simple fix, maybe not. The hacky solution would be to impose a prior on depth of the road immediately in front of the car but that’s not ideal

from dpvo.

Any fix on this issue?

from dpvo.

Thx😀, looking forward to getting your update soon

from dpvo.

@whitesockcat what stride did you use for running on kitti?

from dpvo.

@whitesockcat what stride did you use for running on kitti?
1

from dpvo.

is there any way to solve this scale problem?

Hi @whitesockcat , thanks for sharing the result on KITTI. I am new to KITTI and neural networks. I wonder how you test this algorithm on KITTI datasets. Do you just use KITTI as the input and implement the demo.py as shown in the README file, the same way as TartanAir, like python demo.py --imagedir=<path to image_left> --calib=calib/tartan.txt --stride=1 --viz , what's the format and how you set the calib and stride file? I would be very grateful if you can give me some suggestions.

from dpvo.

is there any way to solve this scale problem?

@zachteed and @whitesockcat,

When I am working with kitti odometry dataset(sequence-05), I also got the similar results, issue with scale. Any fix around? If so, please suggest. Waiting for your reply..

Thank you
Gopi

from dpvo.