This project implements and compares **four neural rendering architectures** on the Tiny NeRF dataset, focusing on improving quality, speed, and training stability. The key innovation is combining **SIREN (Sine-activated networks)** with **Ray Transformers** to achieve state-of-the-art results while maintaining GPU memory efficiency.
computer-vision deep-learning transformers volume-rendering siren attention-mechanism nerf 3d-reconstruction view-synthesis positional-encoding neural-radiance-fields scene-representations implicit-neural-representations
-
Updated
Apr 4, 2026 - Jupyter Notebook