Abstract
Data set is the basis of machine learning, a good data set can promote the development of various applications. Machine learning has been deeply involved in the protection and inheritance of cultural resources. However, there are few data sets about Thangka, and the types and quantity of Thangka images are relatively few. Therefore, we first establish a Thangka data set called RPTK1 (Religious Portrait Thangka Version 1), which contains 3,338 Thangka images, more than any other Thangka data set. The objects in the data set basically cover the common Thangka religious portraits, tools and headdresses, and are marked in the professional language of Buddhism. Then, on the basis of the RPTK1 data set, in order to achieve better detection of small Thangka objects (Thangka religious tools), we propose an improved Single Shot MultiBox Detector (SSD) method, called Single Shot MultiBox Detector with Improvement Feature Fusion And Loss Function (FALSSD). Finally, in order to verify the effectiveness of the FALSSD method, we conduct experiments on the RPTK1 data set. The experimental results show that the mAP of our method in the RPTK1 data set reaches 83.85%. Compared with the other ten state-of-the-art methods, the performance of our model is better.
Highlights
In recent years, machine learning has developed rapidly, from natural scenes to paintings, comics, etc., but the premise is that machine learning needs data sets to support
We propose a new method for object detection in Thangka images based on Single Shot MultiBox Detector (SSD) model
Our experiments show that our FALSSD model has a better detection effect on the religious portraits and headdresses in Thangka images, and It has an impact on smaller objects such as religious tools
Summary
Machine learning has developed rapidly, from natural scenes to paintings, comics, etc., but the premise is that machine learning needs data sets to support. We propose a new work: Thangka data set-RPTK1. The collected data set contains a total of 3338 images of Thangka religious portraits, with a total of 57 labels of Buddhist professional terms. Thangka images in the RPTK1 data set are rich in content, numerous composition elements and various forms of images. Aiming at the poor detection effect of small objects in the RPTK1 data set, we propose a new feature fusion method, and a new positioning loss function to improve the accuracy of final detection. We propose a new method for object detection in Thangka images based on SSD model. Our experiments show that our FALSSD model has a better detection effect on the religious portraits and headdresses in Thangka images, and It has an impact on smaller objects such as religious tools
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.