User Tools

Site Tools


projects:year7:7a.022.tut

7a.022.TUT - Unsupervised Color Constancy using Adversarial Learning

Project - Summary

We propose a novel color constancy approach, called Bag of Color Features (BoCF), building upon Bag-of-Features pooling. The proposed method substantially reduces the number of parameters needed for illumination estimation. At the same time, the proposed method is consistent with the color constancy assumption stating that global spatial information is not relevant for illumination estimation and local information (edges, etc.) is sufficient. Furthermore, BoCF is consistent with color constancy statistical approaches and can be interpreted as a learning-based generalization of many statistical approaches. To further improve the illumination estimation accuracy, we propose a novel attention mechanism for the BoCF model with two variants based on self-attention. BoCF approach and its variants achieve competitive, compared to the state of the art, results while requiring much fewer parameters. Thus, they are more suitable for low computational power devices, e.g., mobile devices.

Project - Team

Team Member Role Email Phone Number Academic Site/IAB
Moncef Gabbouj PI moncef.gabbouj@tuni.fi +358 (400) 736613 Tampere University
Alexandros Iosifidis Co-PI alexandros.iosifidis@tut.fi +45 9350 8875 Tampere University
Jenni Raitoharju Co-PI Jenni.raitoharju@tut.fi +358 50 447 8418 Tampere University
Firas Laakom Student/Researcher firas.laakom@tuni.fi 358 46 5219 0250 Tampere University
Jarno Nikkanen Project Mentor jarnon@xiaomi.com 358 50 483 5323 Intel
Sponsor: Intel

Project - Novelty of Approach

Multilinear techniques have been shown to be competitive with low complexity. The application of multilinear architecture replacing traditional deep architecture is however under-developed. We aim to employ multilinear techniques in network design and pruning to achieve both compactness and performance

Project - Deliverables

Deliverables
1 New robust solution for data scarcity in illumination estimation
2 Comparison with other color constancy approaches
3 Unsupervised color constancy
4 Work on other novel supervised methods

Project - Benefits to IAB

The proposed approach can be used for illumination estimation for color constancy, especially for low computational power devices, e.g., mobile devices.

Project - Presentation Video

Project - Documents

FilenameFilesizeLast modified
7a.022.tut_ip_info_sheet.docx124.4 KiB2019/08/20 11:47
7a.022.tut_final_report.pdf838.9 KiB2019/08/20 09:14
7a.022.tut_speeding_up_neural_networks_quad_2017_fall_meeting.pptx3.4 MiB2019/08/13 15:03
7a.022.tut_quad_chart_2018_spring_meeting.pptx1.8 MiB2019/08/13 15:03
7a.022.tut-unsupervised-color-constancy-executive-summary.pdf108.1 KiB2019/08/13 15:03
7a.022.tut_confluence_project_page.pdf249.2 KiB2019/08/13 15:03
projects/year7/7a.022.tut.txt · Last modified: 2021/06/02 15:09 by sally.johnson