performance issues

[PetropoulakisPanagiotis]

Hi,
I tried to evaluate the performance on the TUM dataset (roo, desk2) and it does not look good.

Has anyone similar experience?

[PetropoulakisPanagiotis]

I found the problem:
The way in dataset.py the rgb and depth frame association is done with make_pair function is wrong.

Please use associate.py from here: https://cvg.cit.tum.de/data/datasets/rgbd-dataset/tools
And use these pairs. Then the code works ok. Not a similar performance as ORBV2 slam or PTAM but is close to ElasticFusion.