Project II: Multimodal Style Transfer
- Multimodal Style Transfer is a hierarchical Deep Convolutional Nueral Network for Fast Artistic Style Transfer.
- This is applied by using Graph Cuts.
- It is based on the assumption that images styles can be described by the deep features present in them.
- Images are scraped using DuckDuckGo and the scraper is applied on GettyImages.
- Then a style is chosen from images that are also scraped based on a text prompt given by the user. The style image is applied on the base image to get the desired image.
- The user is shown a catalogue of style images which are ranked using CLIP. The user can choose from the ranked images to get the best result.
Base Image : Chicken Dinosaur | Style Image : Colourful Feathers
Base Image : Aircraft Carrier | Style Image : Starry Night