' Stylised Shanghai '
Artist: Qiwei Zhang
Kevin Li
Year: Mar 2022
Research: Explore the concepts of "historical recurrence" and "cross-time and space fusion".
Web crawler + Style transfer + Text-to-Video retrieval
The project attempts to explore the concepts of "historical recurrence" and "cross-time and space fusion".
Description
I select the city of Shanghai, China, The image data is selected from the historical changes in different periods.
I attempt to transfer the style of its pictures to pictures of everyday life in London. There will achieve the concept of blending across time and space.
Next, try to generate a video of the transition between the two pictures by the machine to complete the concept of historical reproduction.
Technical Path
2. Generate stylised images:
Use tensorflow's 'Style transfer' framework to uniformly change the style of the picture
Code link: (https://git.arts.ac.uk/20016339/Coding-two/blob/master/Final%20Project/style_video.ipynb
Learning materials: https://www.tensorflow.org/hub/tutorials/tf2_arbitrary_image_stylization
https://www.tensorflow.org/tutorials/generative/style_transfer
1. Select images:
Use a web crawler to download related pictures, then filter 24 images again
Code link: https://git.arts.ac.uk/20016339/Coding-two/blob/master/Final%20Project/spider_img.ipynb
3. Generate stylised video:
Learn additional tensorflow's 'Text-to-Video retrieval' framework to turn images into videos with transition effects
Final watch the video
Code link: https://git.arts.ac.uk/20016339/Coding-two/blob/master/Final%20Project/style_video.ipynb
Learning materials：https://www.tensorflow.org/hub/tutorials/text_to_video_retrieval_with_s3d_milnce
Example:
