Blog posts

Foundation Models (FMs) are transforming Earth Observation (EO) by unifying diverse satellite, environmental, and sensor datasets into powerful multimodal representations. This study surveys the state of the art, covering adaptive architectures, large-scale pretraining pipelines, and generative “any-to-any” frameworks. We examine how these advances are accelerating applications and the integration of EO into digital twins.

Gaussian Splatting is a cutting edge technique for real time neural rendering that models scenes using explicit 3D Gaussians. It offers an alternative to neural implicit representations (NIR) and NeRFs. This study provides a rigorous, structured exploration of its mathematical foundations, differentiable rasterization pipeline, and key advancements that are redefining Gaussian-based scene representation.

This blog offers a comprehensive exploration of Neural Radiance Fields (NeRFs), a method for photorealistic 3D scene reconstruction from sparse 2D images. It covers foundational concepts, training techniques, and notable advancements and well-known variants in the NeRF family.

Welcome to this guide on Diffusion Models, a groundbreaking class of generative models that create high-quality data by refining noisy inputs. This blog covers their foundations, architectures, training, advanced techniques like conditional and latent diffusion, and applications ranging from image editing to medical imaging, offering a concise overview of their impact on machine learning.

Welcome to this guide on State Space Models (SSMs), exploring their efficiency in modeling long-range dependencies, the S4 model, and key techniques.

Welcome to this guide on Neural Implicit Representations (NIR), an advanced approach to 3D reconstruction and graphics. Explore concepts like implicit functions, occupancy networks, volumetric rendering, and Neural Radiance Fields (NeRF), enabling high-resolution 3D modeling.