Qualitative Image Selection with Active Learning
(2023)Department of Automatic Control
- Abstract
- In this work, an active learning pipeline is presented that allows for comparative tests between 3 different types of data scoring methods, uncertainty selection, diversity selection, and loss selection. Tests and parameter sweeps indicate that hyperparameters like reshuffling and early stopping are required to ensure fair comparisons. We then apply our pipeline to test some naive scoring methods on two Computer Vision problems, Image Detection, and Object Detection. We conclude that comparing these scoring methods against random gives minor evaluation loss improvements on our datasets in the right circumstances, and discuss other aspects of data quality along the way.
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9134631
- author
- Allander, Linnea and Nordtorp, Torben
- supervisor
- organization
- year
- 2023
- type
- H3 - Professional qualifications (4 Years - )
- subject
- report number
- TFRT-6198
- other publication id
- 0280-5316
- language
- English
- id
- 9134631
- date added to LUP
- 2023-08-18 15:04:13
- date last changed
- 2023-09-06 13:56:50
@misc{9134631, abstract = {{In this work, an active learning pipeline is presented that allows for comparative tests between 3 different types of data scoring methods, uncertainty selection, diversity selection, and loss selection. Tests and parameter sweeps indicate that hyperparameters like reshuffling and early stopping are required to ensure fair comparisons. We then apply our pipeline to test some naive scoring methods on two Computer Vision problems, Image Detection, and Object Detection. We conclude that comparing these scoring methods against random gives minor evaluation loss improvements on our datasets in the right circumstances, and discuss other aspects of data quality along the way.}}, author = {{Allander, Linnea and Nordtorp, Torben}}, language = {{eng}}, note = {{Student Paper}}, title = {{Qualitative Image Selection with Active Learning}}, year = {{2023}}, }