Issue on Jetson Xavier Nano

[MooreNtu]

Thank you very much for your excellent work on Jetson-SLAM. We attempted to reproduce your reported results using a Jetson Xavier Nano board configured in the 10W power mode.
While we successfully achieved similar accuracy on the EuRoC Machine Hall sequences, we encountered issues achieving the FPS performance mentioned in your paper. Specifically, we're only seeing an average of 18 FPS at a resolution of 752x480, which is below the expected performance.
We are using the same OpenCV version mentioned in your repository.
Could you please suggest if there's anything else we might be missing? Let us know if you need further details to investigate this.
Thanks again!

[ashishkumar822]

Hi. In the paper, we have reported results on Jetson-NX and RTX-2070. For 752x480, Jetson-NX, we achieved 30-35FPS, as mentioned in Figure-1 and Figure-11-c, in the main paper

For Jetson-Nano, It would be less than that of Jetson-NX because Jetson-NX has 384 CUDA cores while Jetson-Nano has only 128 CUDA cores.

Currently, Jetson-SLAM only utilizes CUDA cores.

Hence, your FPS seems good for Nano (128 cores vs 384 cores of Xavier NX).

But let me know what model your board is! Is it orin nano or just nano because in Xavier series, it seems there is no nano variant.

[MooreNtu]

I'm really sorry. Let me correct my mistake. It is not Xavier Nano. It is Xavier NX with 384 CUDA cores.

[ashishkumar822]

I see. Thanks for the information. have you supplied the masks correctly? which variant i.e. stereo/RGBD/Monocular you are running? Can you please provide more context?