Request: 2D Camera Images with Stem Segmentation (from Multi-Camera Payload)

[probstlukas]

Hello,

Thank you for releasing the DigiForests dataset!

For our current project, we're looking for data that supports stem-level instance segmentation. We've been using the GOOSE dataset so far, but its labels for "tree trunk" often include large branches, whereas we're specifically interested in clean stem representations.

Your dataset appears to offer exactly the kind of precise stem segmentation we're looking for -- but currently available only for LiDAR data.

We noticed that your sensor payload includes a multi-camera unit. Would it be possible to provide the corresponding 2D RGB images, along with camera intrinsics and calibration parameters? That would allow us to project the LiDAR-based segmentation into image space for 2D stem detection and training.

If this data is already available and I missed it, I’d be very grateful for a pointer.

Thank you very much for your work and support!

Best regards,
Lukas

[jbehley]

Hi Lukas,

First, thanks for your interest in the data and sorry for the delay.

The camera images are contained in the corresponding ROS bags, but we currently still have to collect the corresponding camera intrinsics/extrinsics. As the backpack had some modifications between the different recordings, we currently have to match the calibrations to the recordings (and verify that these are correct).

We are working on this and will let you know, when we have these recovered and verified. Stay tuned. ;)

[jbehley]

Sorry for the long wait (I have had the code from Nived for some time, but forgot about it :/).

We have the extrinsics now in the folder scripts/calibration/extrinsics; there we have also an example projection with the extracted images and intrinsics. Hope that helps.