/Handling-Color

Primary LanguageJupyter Notebook

Handling Colors in Image Classification

Authors: Soumya Snigdha Kundu, Sheel Patel, Shaswat Srivatsava, Aarsh Chaube and Vaishnavi Moorthy.

Abstract: Self-supervision posses the ability to learn better decision-related representations when compared to that of supervised learning. In this paper, we prove that the representations learnt by a self-supervised model is invariant to any form of color bias. We present two separate experimentation pipelines to evaluate the capability of a model to handle artificial color bias and gauge the ability of a model to incorporate naturally occurring color differences present in vision data.

Primary Results: To be Posted.

Data:

  1. MNIST - LeCun, Y. & Cortes, C. (2010). MNIST handwritten digit database.
  2. Colored MNIST - Kim, B., Kim, H., Kim, K., Kim, S., & Kim, J. (2019). Learning not to learn: Training deep neural networks with biased data. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (pp. 9012-9020).