CSSSketch2Code
Yi Han, Jun He, Qiwen Dong · 2018
With the constantly increasing scale of the Internet and the users, the Internet applications have higher demands on the front-end web pages. Some web pages have single lattice structure and a relatively fixed HTML code template, which can be automatically generated. There have been research abroad using the deep learning on the task of automatically generating the web pages. However due to the basic encoder-decoder model adopted, the generalization ability of the model is not very robust. In this paper, we propose a novel method based on object detection and attention mechanism to automatically generate a web page with CSS style information. We use object detection to extend the original problem, which makes the model possible to detect the CSS style contents in the web page. Meanwhile we use attention mechanism to strengthen the model. Finally we propose our own dataset, based on which the experiment results show that method we proposed outperforms other existing methods.