An Object Detection Method Using Probability Maps for Instance Segmentation to Mask Background

• DOI: 10.20965/jaciii.2023.p0886
• Publication Date: 2023-09-19T15:58:03Z
• Authors: Shinji Uchinoura; Junichi Miyao; Takio Kurita
• Keywords: Foreground detection; Background image

This paper proposes a two-step detector called segmented object detection, whose performance is improved by masking the background region. Previous single-stage detection methods suffer from problem of imbalance between foreground and classes, where occupies more regions in image than foreground. Thus, loss firmly incorporated into training. RetinaNet addresses this with Focal Loss, which focuses on loss. Therefore, we propose method that generates probability maps using instance segmentation first step feeds back generated as masks second prior knowledge to reduce influence enhance We confirm can improve accuracy adding information both input output rather only results. On Cityscapes dataset, our outperforms state-of-the-art methods.