A Dataset for Generating Chinese Landscape Painting

A dataset was constructed for intelligent generation of Chinese landscape paintings based on deep learning, which provides the scientific assistance for inheriting the traditional Chinese culture. This dataset is originated from the two of top ten paintings in ancient China, i.e., “Dwelling in the Fuchun Mountains” with ink wash style and “A Thousand Li of Rivers and Mountains” with style of blue and green. This dataset contains landscape paintings and their corresponding sketches, as well as photos of natural landscapes. The samples of landscape paintings are achieved with the size of $256\times 256$ due to the technique of image preprocessing, the samples of sketches are generated by the canny edge detector, and the samples of landscape photos are formed by image flipping. The above samples of landscape paintings, sketches, and photos make the dataset more flexible for the generation of landscape paintings. Based on this dataset, landscape paintings can be generated from sketches and photos respectively, and landscape paintings can be transferred from one style to another. The experimental results demonstrate the suitability of the dataset for the field of landscape painting generation. The constructed dataset will be publicly available to the community.

I. Introduction

Chinese landscape painting is an important part of traditional Chinese culture [1], and it mainly describes the natural landscape of mountains and rivers. As an ancient expression form of art, landscape painting has unique charm and re-creation value, and it embodies the artistic conception of Chinese painting. With the rapid development of society, economy, and technology, landscape painting forms a unique cultural connotation and artistic expression. It plays an important role in strengthening cultural self-confidence and improving the country's cultural soft power.