Naslov
Multi-domain semantic segmentation with overlapping labels

Autori
Bevandić, Petra ; Oršić, Marin ; Grubišić, Ivan ; Šarić, Josip ; Šegvić, Sinisa

Vrsta, podvrsta i kategorija rada
Radovi u zbornicima skupova, cjeloviti rad (in extenso), znanstveni

Izvornik
Proceeedings of IEEE/CVF Winter Conference on Applications of Computer Vision / Bowyer, Kevin ; Medioni, Gérard ; Scheirer, Walter - Waikoloa (HI) : Institute of Electrical and Electronics Engineers (IEEE), 2022, 2422-2431

ISBN
978-1-6654-0915-5

Skup
IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)

Mjesto i datum
Waikoloa (HI), Sjedinjene Američke Države, 03.01.2022. - 08.01.2022

Vrsta sudjelovanja
Poster

Vrsta recenzije
Međunarodna recenzija

Ključne riječi
computer vision ; semantic segmentation
(računalni vid ; semantička segmentacija)

Sažetak
Deep supervised models have an unprecedented capacity to absorb large quantities of training data. Hence, training on many datasets becomes a method of choice towards graceful degradation in unusual scenes. Unfortunately, different datasets often use incompatible labels. For instance, the Cityscapes road class subsumes all driving surfaces, while Vistas defines separate classes for road markings, manholes etc. We address this challenge by proposing a principled method for seamless learning on datasets with overlapping classes based on partial labels and probabilistic loss. Our method achieves competitive within- dataset and cross-dataset generalization, as well as ability to learn visual concepts which are not separately labeled in any of the training datasets. Experiments reveal competitive or state- of-the-art performance on two multi-domain dataset collections and on the WildDash 2 benchmark.

Izvorni jezik
Engleski

Znanstvena područja
Računarstvo

POVEZANOST RADA