Masking Strategies has been accepted at ICCVW-OODCV 2023
Masking Strategies for Background Bias Removal in Computer Vision Models has been accepted at the OOD-CV workshop at ICCV 2023.
Happy to share that my first PhD paper, titled ‘Masking Strategies for Background Bias Removal in Computer Vision Models,’ co-authored with Cassio F. Dantas, Dino Ienco, and Diego Marcos, has been accepted at the International Conference on Computer Vision (ICCV) 2023 workshop on Out of Distribution Generalization in Computer Vision (OOD-CV) scheduled to be held in Paris in October.
In our research, we focus on fine-grained image classification, such as identifying bird species from images. We identify a common issue where computer vision models learn to associate the species to their habitats (the image background), introducing bias. We propose some simple masking strategies to remove this bias.
- Preprint: ArXiv.
- Code: GitHub
- Published Version: ICCVW-OODCV