Zero-shot_Semantic_NST

Repository for "Zero-shot Semantic Neural Style Transfer for Images", course project for Deep Learning (CS541/Fall 2023).

Steps to run the code:

To perform style transfer on the entire image in a zero-shot manner, use the following command:

python infer.py --content_path CONTENT_PATH --style_path STYLE_PATH --resize --keep_ratio

Example:

python infer.py --content_path data/content/c1.jpg --style_path data/style/candy.jpg --resize --keep_ratio

To perform style transfer on a specific region of the image (semantic segmented mask) using the CLIPSeg Segmentation model, use the following command:

python clipseg_infer.py --content_path CONTENT_PATH --style_path STYLE_PATH --prompts PROMPTS

Example:

python clipseg_infer.py --content_path data/content/parked_car.jpg --style_path data/style/candy.jpg --prompts "car"

A demo of the approach on a chair in the Unity Hall at WPI. The style images are towards the bottom left corner of the resultant images.