Ladder-style DenseNets for Semantic Segmentation of Large Natural Images (CROSBI ID 651152)
Prilog sa skupa u zborniku | izvorni znanstveni rad | međunarodna recenzija
Podaci o odgovornosti
Krešo ; Ivan ; Krapac, Josip ; Šegvić, Siniša
engleski
Ladder-style DenseNets for Semantic Segmentation of Large Natural Images
Recent progress of deep image classification models pro- vides a large potential to improve state-of-the-art perfor- mance in related computer vision tasks. However, the tran- sition to semantic segmentation is hampered by strict mem- ory limitations of contemporary GPUs. The extent of fea- ture map caching required by convolutional backprop poses significant challenges even for moderately sized PASCAL images, while requiring careful architectural considera- tions when the source resolution is in the megapixel range. To address these concerns, we propose a DenseNet-based ladder-style architecture which is able to deliver high mod- elling power with very lean representations at the original resolution. The resulting fully convolutional models have few parameters, allow training at megapixel resolution on commodity hardware and display fair semantic segmenta- tion performance even without ImageNet pre-training. We present experiments on Cityscapes and Pascal VOC 2012 datasets and report competitive results.
semantic segmentation
nije evidentirano
nije evidentirano
nije evidentirano
nije evidentirano
nije evidentirano
nije evidentirano
Podaci o prilogu
1-8.
2017.
objavljeno
Podaci o matičnoj publikaciji
Proceedings of the IEEE International Conference on Computer Vision Workshop
Ikeuchi, Katsushi ; Medioni, Gérard ; Pelillo, Marcello
Venecija: Institute of Electrical and Electronics Engineers (IEEE)
Podaci o skupu
IEEE International Conference on Computer Vision Workshop
poster
23.10.2017-29.10.2017
Venecija, Italija