Specular Highlight Semantic Dataset

Citation Author(s):
Leyun
Sun
Jing
Chen
Submitted by:
Jing Chen
Last updated:
Fri, 03/28/2025 - 03:22
DOI:
10.21227/v1a0-jd82
License:
0
0 ratings - Please login to submit your rating.

Abstract 

The Semantic Highlight Mask Dataset holds significant importance in the field of computer vision, particularly for tasks such as object detection, image segmentation, and scene understanding. This paper proposes a novel Semantic Highlight Mask Dataset designed to support high-precision semantic segmentation and emphasize key regions in images. The dataset encompasses a diverse range of image categories, with finely annotated highlight masks assigning semantic labels and saliency weights to each pixel. The dataset construction process integrates automated algorithms with manual validation to ensure annotation quality and consistency. Experimental results demonstrate that this dataset significantly enhances the performance of deep learning models in semantic segmentation tasks, particularly when handling complex backgrounds and fine-grained targets. We further explore the dataset’s potential in real-world applications, such as autonomous driving, medical image analysis, and augmented reality, providing valuable resources and benchmarks for future research.

Instructions: 

This dataset covers seven object types: Bolt, Column, Loop, Pedestal, Slice, Stick, and Wheel. Each object type includes four lighting conditions: ceiling, point, spotlight, and sunlight. The images consist of original images, highlight masks prefixed with "gt_", and object masks prefixed with "ob_". The file "feature.xlsx" contains some provided semantic features, which are converted into "text_test" and "text_train" file formats for semantic input into the network.