Developing a SOTA Autoencoder model backed by physics based loss functions to generate alternate realities froma single stationary point in time.
Developed a physics-aware deep learning framework for detecting mirage-like optical distortions in natural scenes using depth estimation, flow modeling, and segmentation.
Developed an Image inpaining model that fills in the irregular, random shaped masks (inspired by forensic scenarios).
Built a multimodal architecture combining Faster R-CNN, CLIP encoders, and a GPT-based decoder for radiology report generation from chest X-ray images.