Leveraging Deep-learning for Image editing and 3D Reconstruction

Deep learning-based techniques provide a powerful framework to significantly simplify visual content creation, including images, videos, or 3D objects, by producing convincing results from intuitive, high-level user input. I will discuss several techniques from our recent research for sketch-based image editing and image-based 3D reconstruction, and provide some background on methods including conditional GANs and differentiable rendering that enable us to achieve this. Our techniques allow users to edit and create convincing content including photorealistic images and 3D objects with minimal manual modeling efforts.