人工智能
变压器
计算机科学
计算机视觉
图像分辨率
高分辨率
融合
特征(语言学)
低分辨率
模式识别(心理学)
地质学
遥感
工程类
电压
哲学
电气工程
语言学
作者
Chao Yao,Shuaiyong Zhang,Mengyao Yang,Meiqin Liu,Junpeng Qi
标识
DOI:10.1109/icme51207.2021.9428393
摘要
Depth maps have been still suffering from some non-negligible effects, resulting from the consumer-level sensors. The limited resolution of the acquired depth maps is one of these annoying issues. Many prominent researchers have recently made a lot of efforts, such as traditional filters, as well as the deep learning paradigms. However, depth super-resolution is still an open challenge. In this paper, we design a texture-depth transformer for depth super-resolution task, which can learn the corresponding structural information of the high-resolution texture images and the corresponding interpolated depth maps. Moreover, a multi-scale feature fusion strategy is exploited to further enhance the fusion feature. Complementary to a quantitative evaluation, we demonstrate the effectiveness of the proposed approach.
科研通智能强力驱动
Strongly Powered by AbleSci AI