150,795 synthetic household images for continual learning research
HOWS-CL-25 is a synthetic dataset created with BlenderProc, containing 150,795 unique images across 25 household object categories built from 925 3D models.
Every RGB image ships with a matching depth, segmentation, and normal image, and the images are organized into five sequences for continual learning evaluation — the setting that motivated RECALL.
Zenodo, 2022
@inproceedings{knauer2022hows,
title = {HOWS-CL-25: Household Objects Within Simulation Dataset for Continual Learning},
author = {Knauer, M. and Denninger, M. and Triebel, R.},
booktitle = {Zenodo},
year = {2022},
doi = {10.5281/zenodo.7189434}
}