Interactive image selection for user annotation aided by the feature space projection

• Grant number: 21/06545-5
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Start date: July 01, 2021
• End date: June 30, 2022
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: Alexandre Xavier Falcão
• Grantee: Gabriel Dourado Seabra
• Host Institution: Instituto de Computação (IC). Universidade Estadual de Campinas (UNICAMP). Campinas , SP, Brazil
• Associated research grant: 14/12236-1 - AnImaLS: Annotation of Images in Large Scale: what can machines and specialists learn from interaction?, AP.TEM

Abstract

The development of machine learning models based on convolutional neural networks over the years has contributed to important advances in image classification and segmentation problems. Many of these models, however, still rely on a large number of labeled samples for training, which may require considerable user effort and time. Such cost may be prohibitive in areas such as Biology and Medicine. In this sense, the need for developing weakly supervised learning alternatives becomes evident - i.e., learning methods capable of using a few labeled samples to create effective models. Feature Learning from Image Markers (FLIM) fits into this context, as it only uses information about very few samples, with user-marked regions of interest, to extract features that will ultimately be used for training an image classifier. However, there are still unanswered questions regarding FLIM: How should we select the best samples and regions for drawing markers? How can we use the visual impact of these choices in the architectural design of the FLIM extractor? To answer these questions, an interactive interface will be developed that allows image and marker selection to improve the feature space. Such improvement should be evident in the 2D projection of the feature space. The workflow will be validated in the context of image classification and compared to a popular extractor model based on an autoencoder network. (AU)