ueda0319 / imap_pytorch Goto Github PK

Hello, when I export the pose data and use the tum script 'evaluate_ate.py' for evaluation, I cannot get the correct result. How to save and evaluate the pose data?

Hi,
I just wanted to say thanks for sharing your code! It was really helpful to learn NERF and iMAP. While learning the algorithm through the paper and your code I did a bunch of refactoring to make variable names clear, added comments, etc. I don't expect this is useful for you as the author, but in case you might be interested I thought I'd share the fork:

https://github.com/originrose/imap_slam_torch

Is there a possibility to use live depth stream from for example realsense camera?

Hi,
I'm running this with Pytorch 1.8 and Cuda 11.1 on a 3090 and I'm getting around 0.75 FPS. In the paper they claim 10 for the tracking thread and 2 for mapping. Is this a problem on my side or normal for your implementation?
Also, my rendered images also seem to be much worse than in your gif:

Do you know what the problem could be here?

I'm running run_slam_tum.py on fr1/teddy