Single camera based unsupervised and self-supervised depth estimation.
Our project explores monocular depth estimation using Unsupervised and Self-Supervised methods. Particularly, we explore and replicate two networks, Monodepth2 and HRdepth which tackle depth estimation as an image reconstruction problem. Both networks attempt to provide more accurate and efficient networks for depth estimation for the self-driving car industry.
The project tech stack: Python, PyTorch, AWS Sagemaker, AWS S3 bucket, Flask API, AWS ElasticBeanStalk. Produced a deployed endpoint that would allow to upload image/video file and produce depth disparity heat maps ( lighter the color, closer the object).
