Representation Learning by Learning to Count

Noroozi, Mehdi; Pirsiavash, Hamed; Favaro, Paolo (2017). Representation Learning by Learning to Count. In: 2017 IEEE International Conference on Computer Vision (ICCV). IEEE

[img] Text
Noroozi_Representation_Learning_by_ICCV_2017_paper.pdf - Published Version
Restricted to registered users only
Available under License Publisher holds Copyright.

Download (1MB) | Request a copy

We introduce a novel method for representation learning that uses an artificial supervision signal based on count- ing visual primitives. This supervision signal is obtained from an equivariance relation, which does not require any manual annotation. We relate transformations of images to transformations of the representations. More specifically, we look for the representation that satisfies such relation rather than the transformations that match a given representation. In this paper, we use two image transformations in the context of counting: scaling and tiling. The first transformation exploits the fact that the number of visual primitives should be invariant to scale. The second trans- formation allows us to equate the total number of visual primitives in each tile to that in the whole image. These two transformations are combined in one constraint and used to train a neural network with a contrastive loss. The pro- posed task produces representations that perform on par or exceed the state of the art in transfer learning benchmarks.

Item Type:

Conference or Workshop Item (Paper)


08 Faculty of Science > Institute of Computer Science (INF) > Computer Vision Group (CVG)
08 Faculty of Science > Institute of Computer Science (INF)

UniBE Contributor:

Noroozi, Mehdi and Favaro, Paolo


000 Computer science, knowledge & systems
500 Science > 510 Mathematics






Xiaochen Wang

Date Deposited:

20 Apr 2018 12:43

Last Modified:

20 Apr 2018 12:43




Actions (login required)

Edit item Edit item
Provide Feedback