A. Alvarez-Gila, J. van de Weijer, Y. Wang, and E. Garrote, “MVMO: A Multi-Object Dataset for Wide Baseline Multi-View Semantic Segmentation,” ICIP 2022.
Abstract:
MVMO is a synthetic dataset for wide baseline multi-view semantic segmentation. The dataset comprises 116,000 scenes with randomly placed objects from 10 distinct classes, captured from 25 camera locations in the upper hemisphere. It features photorealistic, path-traced image renders with semantic segmentation ground truth for every view. Unlike existing multi-view datasets, MVMO is characterized by wide baselines between cameras, high density of objects, large disparities and heavy occlusions, and view-dependent object appearance. The dataset aims to advance research in multi-view semantic segmentation and cross-view semantic transfer by addressing challenges that single-view semantic segmentation faces due to self and inter-object occlusions. The authors provide baselines demonstrating that new research is needed to effectively exploit the complementary information available in multi-view setups.
Bibtex:
@inproceedings{alvarez_gila_mvmo_icip_2022,
title = {MVMO: A Multi-Object Dataset for Wide Baseline Multi-View Semantic Segmentation},
booktitle = {2022 IEEE International Conference on Image Processing (ICIP)},
year = {2022},
address = {Bordeaux, France},
author = {{Alvarez-Gila}, Aitor and van de Weijer, Joost and Wang, Yaxing and Garrote, Estibaliz},
doi = {10.1109/ICIP46576.2022.9897955},
month = oct
}