A Random CNN Sees Objects: One Inductive Bias of CNN and Its Applications
DOI:
https://doi.org/10.1609/aaai.v36i1.19894Keywords:
Computer Vision (CV)Abstract
This paper starts by revealing a surprising finding: without any learning, a randomly initialized CNN can localize objects surprisingly well. That is, a CNN has an inductive bias to naturally focus on objects, named as Tobias ("The object is at sight") in this paper. This empirical inductive bias is further analyzed and successfully applied to self-supervised learning (SSL). A CNN is encouraged to learn representations that focus on the foreground object, by transforming every image into various versions with different backgrounds, where the foreground and background separation is guided by Tobias. Experimental results show that the proposed Tobias significantly improves downstream tasks, especially for object detection. This paper also shows that Tobias has consistent improvements on training sets of different sizes, and is more resilient to changes in image augmentations.Downloads
Published
2022-06-28
How to Cite
Cao, Y.-H., & Wu, J. (2022). A Random CNN Sees Objects: One Inductive Bias of CNN and Its Applications. Proceedings of the AAAI Conference on Artificial Intelligence, 36(1), 194-202. https://doi.org/10.1609/aaai.v36i1.19894
Issue
Section
AAAI Technical Track on Computer Vision I