Developed a novel approach for segment-based neural style transfer, combining AdaIN layers for real-time style transfer with the Segment Anything model for accurate segmentation
Implemented an interactive user interface, enabling user-guided selection of content, style, target regions, and style loss weights for creative exploration
The project offered a user-centric solution for artistic image manipulation, surpassing the limitations of traditional methods by providing localized style control
Developed and trained a convolutional autoencoder for accurate image colorization of black and white photographs using Python and PyTorch
Trained the model on nearly 28,000 images real-world images from the Google Landmarks dataset