Color Representation in Deep Neural Networks

Published in ICIP, 2017


Convolutional neural networks are top-performers on image classification tasks. Understanding how they make use of color information in images may be useful for various tasks. In this paper we analyze the representation learned by a popular CNN to detect and characterize color-related features. We confirm the existence of some object- and color-specific units, as well as the effect of layer-depth on color-sensitivity and class-invariance.

Paper Poster


If you found this work useful, please cite the associated paper:

M. Engilberge, E. Collins, and S. Susstrunk, “Color representation in deep neural networks,” in Proceedings of the IEEE International Conference on Image Processing, Sep. 2017, pp. 2786–2790


  title = {Color Representation in Deep Neural Networks},
  booktitle = {Proceedings of the IEEE International Conference on Image Processing}
  author = {Engilberge, Martin and Collins, Edo and Susstrunk, Sabine},
  year = {2017},
  pages = {2786--2790}