search

ika-rwth-aachen / MultiCorrupt

MultiCorrupt: A benchmark for robust multi-modal 3D object detection, evaluating LiDAR-Camera fusion models in autonomous driving. Includes diverse corruption types (e.g., misalignment, miscalibration, weather) and severity levels. Assess model performance under challenging conditions.
MIT License
32 stars 4 forks source link

Fewer generated foggy images in nuscenes-mini dataset #6

Closed Neko-coffee closed 1 month ago

Neko-coffee commented 1 month ago

I followed the steps to generate foggy weather data using the nuscenes-mini dataset, but the number of generated images is significantly less than the original dataset. For example, the number of images in the CAM_BACK folder has decreased from 404 (original) to 81 (foggy). I would appreciate any insights or suggestions on why this might be happening and how I can resolve it. Thank you!

TillBeemelmanns commented 1 month ago

Hi, two possible answers:

Bests, Till

Neko-coffee commented 1 month ago

Hi, two possible answers:

  • Only images from the validation split are converted. Images from the training set are not considered.
  • For camera images, only the samples are converted, but not the sweep camera images

Bests, Till

Dear Till,

I'm a newbie with some basic questions, thanks for your patience! Based on your reply, does the image conversion depend on the file nuscenes_infos_val.pkl? Therefore, if I want to convert all the images, should I run the process again using nuscenes_infos_train.pkl? This way, I can convert all the images, or is there another method available?

TillBeemelmanns commented 1 month ago

Yes, the conversion depends on the filepathes in the nuscenes_infos_val.pkl. You can try to use nuscenes_infos_train.pkl, I have never tested it though.

You could also try to adapt the conversion scripts and parse the filepathes after your liking, and then do the conversion.

Neko-coffee commented 1 month ago

Yes, the conversion depends on the filepathes in the nuscenes_infos_val.pkl. You can try to use nuscenes_infos_train.pkl, I have never tested it though.

You could also try to adapt the conversion scripts and parse the filepathes after your liking, and then do the conversion.

Thank you once again for taking the time out of your busy schedule to help me resolve the issue.

Your support is greatly appreciated.