Neural Field
Neural fields represent data as continuous functions parameterized by neural networks, aiming to efficiently and accurately model complex, high-dimensional data. Current research focuses on improving training speed and efficiency through novel initialization schemes and architectural designs, such as sinusoidal neural fields and multi-resolution approaches incorporating techniques like diffusion nets and Fourier features. These advancements are impacting diverse fields, enabling applications in 3D scene reconstruction, medical imaging (e.g., cardiac modeling, photoacoustic tomography), fluid dynamics simulation, and robotics, by offering flexible, continuous representations that surpass traditional methods in accuracy and efficiency.
166papers
Papers
February 20, 2025
February 15, 2025
February 7, 2025
February 4, 2025
January 24, 2025
January 15, 2025
January 6, 2025
December 30, 2024
December 23, 2024