search

Hongbin98 / MonoTTA

Code release for the ECCV 2024 paper 'Fully Test-Time Adaptation for Monocular 3D Object Detection'
MIT License
24 stars 2 forks source link

Could you provide us with the complete KITTI-C dataset? #4

Closed heegwang27 closed 2 months ago

heegwang27 commented 2 months ago

It seems like the data only includes Gaussian corruption. Could you provide us with the complete KITTI-C dataset?

Hongbin98 commented 2 months ago

Thanks for your attention to our work. All datasets in this paper will be public in the future~

However, I am still looking for a reliable way to share these files (~= 4~6GB * 13) with my readers due to the limitation of Google Drive.

It is also worth mentioning that you can prepare the kitti-c dataset yourself by following this link: https://github.com/hendrycks/robustness.

heegwang27 commented 3 weeks ago

Thank you for your response. I look forward to accessing the data in the future.

I have a question though—when I look at Figure 1 in the supplementary material of the paper, the KITTI-C dataset seems to appear as BGR images.

Is there a specific reason for this? Additionally, why is only the "pixelate" corruption in RGB format?

Hongbin98 commented 3 weeks ago

You can check this link for the data 'We have released the full KITTI-C dataset in huggingface! Please check the link for more: https://huggingface.co/datasets/anthemlin/kitti-c' :)

For your doubt of the "pixelate" corruption, you may check this paper 'Benchmarking Neural Network Robustness to Common Corruptions and Perturbations' for more details.

heegwang27 commented 3 weeks ago

Thank you very much for your kindness. Thanks to your help, I was able to download and check the data. However, I still have a question: it seems that the R and B channels in the KITTI-C dataset have been swapped. For example, a red car in the original dataset appears blue in this dataset. Is there a specific reason for swapping the R and B channels? I couldn't find any mention of this in the paper you referred to. Thank you :)

Hongbin98 commented 3 weeks ago

I see what you mean, I will check the generated data and correct how the image is saved (One possible reason is that the default setting of opencv-python does not meet my requirements). Thank you for kindly pointing out this issue~ I will update the kitti-c dataset once this minor issue is solved.