DeepSegRock

Since Google Drive and Zenodo have very limited storage space, our dataset requires approximately 200GB of storage space, so we uploaded the dataset to BaiduNet. The relevant synchronization instructions are also published on the GitHub platform (https://github.com/ZhangGroup-JLU/DeepSegRock). We have released the segmentation benchmark, namely DeepSegRock, which includes two datasets: the Linear Rock Mass Discontinuity Segmentation Dataset (LRMDSD) and the Planar Rock Mass Discontinuity Segmentation Dataset (PRMDSD). If you are in China, it will be more convinient to download with BaiduNet. Below are the download links and extraction codes for the datasets. If you have any questions, please feel free to contact me at [e-mail] or [e-mail]. Website Link: https://pan.baidu.com/s/1S8DdemL1CbE7mrZgGSXiMQ Extract Code: 66ps For visiting the Baidu Net on a windows computer, one can access the following website: https://www.baidu.com/link?url=Vc9deB4dvzPYiItTBh7xriHpna2RuYUMfEYfjZ7o03y&wd=&eqid=fdf7d5180472f68f0000000667511c70 The segmentation benchmark of the paper "DeepSegRock: A Pixel-Level Segmentation Benchmark for Rock Mass Discontinuity Recognition With Deep Learning Architectures" Due to the critical significance of rock mass discontinuities in controlling slope stability, they have attracted considerable attention in precisely recognizing such geological features. Recently, researchers have employed numerous deep learning algorithms to advance the study of rock mass discontinuity recognition. However, these algorithms are primarily evaluated using synthetic datasets or a limited number of selected real-world images, leaving unclear how they perform on images captured in complex practical scenarios and how to measure research progress in this field. To bridge this research gap, a pixel-level segmentation benchmark for rock mass discontinuity is proposed in this paper, namely DeepSegRock, which includes the Linear Rock Mass Discontinuity Segmentation Dataset (LRMDSD) and the Planar Rock Mass Discontinuity Segmentation Dataset (PRMDSD). This benchmark presents the comprehensive perceptual study and analysis of rock mass discontinuity segmentation using large-scale real-world images of high-steep rock slopes. LRMDSD contains over 300, 000 linear rock mass discontinuities, PRMDSD includes more than 320, 000 planar rock mass discontinuities. Based on this segmentation benchmark, a comprehensive qualitative and quantitative deep learning study of current mainstream rock mass discontinuity segmentation algorithms is conducted. These deep learning models trained using the proposed segmentation benchmark in this study offer a new solution for acquiring rock mass discontinuity parameter information in engineering geology, thereby promoting the development of fine-grained automatic interpretation of rock mass discontinuities.