HKUST-Aerial-Robotics/OmniNxt

I am having trouble reproducing your depth pointcloud results.

akira-akp opened this issue · 6 comments

akira-akp commented

Thank you for your repository!

I have a problem when I run the code.
I think that I I think that by using this repository and this dataset, I can get depth pointcloud and SLAM trajectories like in this video(https://www.youtube.com/watch?v=IOuJ7Y6dpeY).

However, when I run it, I get the following pointcloud results, which are not what I am able to reproduce.

Is this the correct combination of rosbag recording data and calibration data I should be using?
If it is incorrect, I would like to know the correct combination.
Additionally, I would like to know if there are any changes in the source code that need to be made to reproduce the data.

Also, is it correct that the aruducam photos in the rosbag recording data are upside down?

branch: pr_fix_main(https://github.com/HKUST-Aerial-Robotics/D2SLAM.git)
model file: hitnet_1x240x320_model_float16_quant_opt.onnx
engine file: hitnet_1x240x320_model_float16_quant_opt.trt

data set DL link: https://hkustconnect-my.sharepoint.com/:f:/g/personal/pliuan_connect_ust_hk/EhrnbPJoptRGvYhsa3P6KUIBscF9NArlttBNUqCaTvzsyw?e=QChj0X
using calibration file: quad_cam_calib-camchain-imucam-7-inch-n3.yaml

rosbag data: quadcam_7inch_n3_2023_1_14/eight_noyaw_1-sync.bag
image_depthPC_7inch

rosbag data: quadcam_ri_sep/quadwalk_around_2-sync.bag
image_depthPC_ri

Peize-Liu commented

Hi,
Sorry for the late reply. The image sequence of OmniNxt is different from D2SLAM, which I think is the main reason for your issue. I will provide a new calibration file and OmniNxt dataset this week.

akira-akp commented

Thanks for your reply.
I am very much looking forward and waiting for the new calibration files and OmniNxt dataset to be provided by you.
I would like you to provide them via OneDrive as I cannot download them via Baidu or Dropbox in my environment.

akira-akp commented

Hello. @Peize-Liu
I am looking forward to the new calibration files and OmniNxt dataset.
When will these be available?

akira-akp commented

Hello. @Peize-Liu

I was able to output my expected point cloud using “quadcam_7inch_n3_2023_1_14/eight_noyaw_1-sync.bag” and “quadcam_ri_sep/quadwalk_around_2-sync.bag”.

The reason for the incorrect point cloud output as in my previous post was due to the following items.

  • The camera indexes in the calibration file are incorrect.
  • The aruducam photos in the rosbag recording data are upside down after rectify processing.

I think that the difference in image sequence between D2SLAM and OmniNxt has affected this part of the process.
I would like to hear your opinion if this idea is correct!

Also, I continue to look forward to receiving new calibration files and OmniNxt datasets from you.

Best regards.

892692498 commented

Hello， @Peize-Liu .
I am also looking forward to receiving new calibration files and OmniNxt datasets from you. Thank you very much.
Best regards.