StyleNet: Generating Attractive Visual Captions with Styles

* under development

• Powered by DLHacks

StyleNet is a novel framework to address the task of generating attractive captions for images and videos with different styles. A novel model component, named factored LSTM is used in StyleNet, which automatically distills the style factors in the monolingual text corpus.

framework

examples of generated captions

Description

• A pytorch implemention of StyleNet
• Author: Chuang Gan, Zhe Gan, Xiaodong He, Jianfeng Gao, Li Deng
• Published in: Computer Vision and Pattern Recognition (CVPR), 2017
• URL: https://www.microsoft.com/en-us/research/wp-content/uploads/2017/06/Generating-Attractive-Visual-Captions-with-Styles.pdf
• Dataset: https://zhegan27.github.io/Paper.html
• Slideshare: https://www.slideshare.net/DeepLearningJP2016/dl-hacks-stylenet-generating-attractive-visual-captions-with-styles
• written by Kota Kakiuchi

Requirement

• python 3.5.3
• pytorch 0.2.0
• torchvision 0.1.9
• numpy 1.13.3
• scikit-image 0.13.1
• nltk 3.2.5