Request: Camera and PCD pairs

[adcosta5]

Hello,

Thank you for releasing the DigiForests dataset!

For our current project, we're looking for image and point clouds pairs for training a forest navigation model.

From the website I was able to download all the pcds and labels. I tried to extract the images from the rosbags, but the timestamps do not match. Looking in the repo I found the calibration_example.zip which contain the pairs. I was wondering if it could be possible to have access to the rest of pairs from the dataset. Or if I've missed and it already exists or there is a super quick way to get them you could point me to it.

Thank you very much for your work and support!

Best regards,
Adrià

[jbehley]

Thanks for your interest in the dataset.

First, it's correct that one has to take the closest image to the corresponding LiDAR scans to perform the projection; the setup is not perfectly synchronized as this is always an issue with 360° LiDARs; one has to essentially take the closes timestamp and then project the points around the acquistion time of the image.

Second, for a better alignment between images and LiDAR, on has to even account for the movement of the LiDAR sensor; this is usally achieved via some LiDAR odometry/SLAM approach. In our case, it was the VILENS system. As our dataset mainly is about the semantics, we now only provide semantics for the already integrated scans.

We are currently looking into preparing a pipeline based on RKO-LIO (paper available at https://arxiv.org/pdf/2509.06593), code is available at https://github.com/PRBonn/rko_lio that can produce the provide the point clouds accounting for the movement of the LiDAR sensor exploiting inertial information. However, it will take a bit.