OPEDD and OFFSED are a set of outdoor off-road stereo image datasets:

OPEDD

• 1018 rectified stereo RGB images with depth from stereo
• Instance masks and IDs for all persons in the left images
• 5 environments: meadows, woods, construction sites, farmland and paddocks
• Wide variety of poses, e.g. extended limbs, handstands, crouching or lying down

OFFSED

• Full semantic segmentation annotations for 203 images from OPEDD
• 19 classes: grass, trees, sky, drivable and non-drivable dirt, obstacles, crops, building, person, bush, wall, drivable and non-drivable pavement, carried object, truck, car, excavator, guard rail and camper

Download

Publications:

OFFSED: Off-Road Semantic Segmentation Dataset (Abstract & PDF)

Please cite:

P. Neigel, J. R. Rambach, and D. Stricker, "OFFSED: Off-Road Semantic Segmentation Dataset," in Proc. of the 16th Int. Conference on Computer Vision Theory and Applications - Volume 4: VISAPP, pp. 552-557, 2021

OPEDD: Off-Road Pedestrian Detection Dataset (Abstract & PDF)

Please cite:

P. Neigel, M. Ameli, J. S. Katrolia, H. Feld, O. Wasenmüller, and D. Stricker, "OPEDD: Off-Road Pedestrian Detection Dataset," in Journal of the Int. Conference in Central Europe on Computer Graphics, Visualization and Computer Vision (WSCG), 28(1-2), pp. 207-212, 2020

© 2021