search

EasonChen99 / I2D-Loc

Official code for "I2D-Loc: Camera Localization via Image to LiDAR Depth Flow"
47 stars 6 forks source link

Pretrained Model Performance #6

Open cattaneod opened 1 year ago

cattaneod commented 1 year ago

Thank you for releasing the code of your work.

I'm trying to reproduce the results reported in your paper with the provided pretrained model. However, I'm getting worse performance on KITTI:

Mean trans error 46.04930  Mean rotation error 1.17001
Median trans error 30.49748  Median rotation error 0.81869
Outliers number 269/4541

And on Argoverse the performance is even worse.

Mean trans error 104.11842  Mean rotation error 2.61865
Median trans error 79.57483  Median rotation error 1.88286
Outliers number 610/2740

How can I reproduce the results reported in the paper?

Thank you.

EasonChen99 commented 1 year ago

I apologize for the delay in my response.

I've just downloaded the program and tested it using the pretrained model. Here are the results I obtained on KITTI:

Mean trans error 24.53336  Mean rotation error 0.71082
Median trans error 17.46164  Median rotation error 0.53232
Outliers number 86/4541  Mean 0.13913 per frame

From the results, I believe there are no issues with the uploaded code.

Regarding performance on Argoverse, please note that the pretrained model was exclusively trained on KITTI. As a result, it may not achieve a performance similar to what was reported in the paper when applied to Argoverse.