Thangka Image Segmentation Method Based on Enhanced Receptive Field

Abstract

The portrait thangka image is a kind of religious scroll painting that expresses figures' identity and duties through portraits, sitting platforms, and backlighting. The segmentation of significant semantic objects in the image is one of the essential ways for scholars to study and understand the image's content. To better understand this content, we elaborately collected a dataset of portrait-like thangkas, which consists of 4086 images covering four object categories. We provide rich annotation for this dataset. In addition, we propose an end-to-end deep learning method that effectively solves the problems of blurred target edges, segmentation errors, and missed segmentation in thangka image segmentation. First, regular convolution and atrous convolution of different sizes are concatenated after the high-level feature output. This method can effectively improve the receptive field of the model while obtaining more image feature information. Then, the attention module is introduced to fully utilize the spatial relationship between the image's semantic content and enhance the discriminative ability of the feature representation on the thangka image. Finally, cross-layer feature fusion is added to reduce the loss of edge details and improve the accuracy of target edge segmentation. The results show that compared to the base model, the mPA and mIoU indices of the model proposed in this paper reach 90.75% and 85.66%, respectively, which effectively improved the accuracy of the Tangka image segmentation.

YOLOv8-Mu: An Improved YOLOv8 Underwater Detector Based on Large Kernel Block and Multi-branch Heavy Parameterization Module

“…Deep Convolutional Neural Networks (CNNs) have demonstrated outstanding capabilities in handling complex visual tasks, where adjusting parameters such as network depth and convolutional kernel size to modulate the network's receptive field has become a common strategy for improving prediction accuracy. This is particularly crucial in applications requiring dense predictions such as semantic image segmentation [5] [6], stereo vision analysis [7], and optical flow estimation [8], as these tasks rely on a comprehensive understanding of the extensive context surrounding each pixel to ensure no critical information is overlooked. In this study, we adopted the innovative LarK Block from UniRepLKNet [9], which extends the model's receptive field by leveraging large kernel blocks without the need to increase network layers, effectively enhancing the network's ability to capture details.…”

“…The Dilated Reparam Block is proposed based on equivalent transformation, aiming to enhance feature extraction by combining a non-sparse large-kernel convolutional layer with multiple sparse small-kernel convolutional layers. The key hyperparameters of this method include the size of the large kernel K, the size of parallel convolutional layers k, and the sparsity rate r. Assuming there are four parallel layers with K=9, r=(1,2,3,4), and k= (5,3,3,3). To utilize a larger K, more layers can be enhanced by increasing the kernel size or expanding the sparsity rate.…”