CapText-Unet: Efficient Scene Text Detection using UNET

Scene text detection in natural images remains a challenging task due to variations in text appearance, orientation, and background complexity. This paper proposes a novel approach, CapText-Unet, which leverages the U-Net architecture to detect text regions in scene images. Our method adapts U-Net to predict heatmaps representing text regions, using bounding box annotations to generate pseudo-segmentation labels. The proposed approach achieves state-of-the-art results on the MSRA-TD500 dataset, demonstrating its effectiveness in detecting text regions with high accuracy. The CapText-Unet (CaptureText-UNet) architecture is simple, efficient, and requires only bounding box annotations, making it a promising solution for scene text detection applications.