WHU-USI3DV / CoFiI2P

[IEEE RA-L 2024 & ICRA'25] CoFiI2P: Coarse-to-Fine Correspondences-Based Image-to-Point Cloud Registration
https://whu-usi3dv.github.io/CoFiI2P/
92 stars 5 forks source link

error when run evaluation #12

Closed YhQIAO closed 2 months ago

YhQIAO commented 2 months ago

after proprocessing unscene test dataset and run evaluation.py and get this error image

YhQIAO commented 2 months ago

image it seems that the dimension of two tensor dont match

YhQIAO commented 2 months ago

image after i change the number in model/network.py line 18 and 19(opt.img_H / 32 ...) original is 8, it works but i wonder whether there are some mistakes in tensor dimension and it may lead to wrong result?

YhQIAO commented 2 months ago

Running train. py also has an issue on this line, with a tensor dimension mismatch error

martin-liao commented 2 months ago

Thanks for your interest. I am checking the code now.

martin-liao commented 2 months ago

image after i change the number in model/network.py line 18 and 19(opt.img_H / 32 ...) original is 8, it works but i wonder whether there are some mistakes in tensor dimension and it may lead to wrong result?

Yes, the modification would lead to a wrong result.

martin-liao commented 2 months ago

I have checked the evaluation code. It seems that everything is right. Screenshot from 2024-09-07 10-50-09 Could you please provide the processed data? An incorrect input data shape might cause the issue.

YhQIAO commented 2 months ago

thank you,i will try to reprocess data again and try evaluation again

YhQIAO commented 2 months ago

I tried processing the data again and running it again, but I still get the exact same error message. Could it be that some parameters have been modified during Nuscene's data preprocessing (e.g. options file)?

YhQIAO commented 2 months ago

image the dimension of processed nuscene images and pointcloud is shown in image

martin-liao commented 2 months ago

Ohhhhh... Image size should be (H, W)=(320, 640) instead of (160, 320). I will check the data pre-processing code.

martin-liao commented 2 months ago

Please set the img_scale to 0.4 instead of 0.2 here. I will update the code later.

martin-liao commented 2 months ago

If you still have any questions, please feel free to contact me.

YhQIAO commented 2 months ago

Thank you for your patience, I will close this issue after I try to run it successfully.