Depth in convolutional neural networks solves scene segmentation
Noor Seijdel, Nikos Tsakmakidis, Edward H.F. de Haan, Sander M. Bohté, H. Steven Scholte
This paper shows that deeper convolutional neural networks better distinguish objects from backgrounds in scenes, suggesting that scene segmentation can emerge from network depth without explicit mechanisms.
PLoS Computational Biology · Jul 24, 2020Peer-reviewedNeuroscience