https://giacolees.github.io/tags/computer-vision/Recent content in Computer Vision on giacolees - Tech BlogHugoenSat, 16 May 2026 12:32:26 +0200https://giacolees.github.io/posts/radiance-fields/Sat, 16 May 2026 12:32:26 +0200https://giacolees.github.io/posts/radiance-fields/TL;DR Every scene has a plenoptic function describing every ray of light passing through it — radiance fields are just different ways to approximate and query it. Vanilla NeRF bakes the whole scene into an MLP and integrates densities along each ray, which is data-efficient but brutally slow (~78 billion MLP evaluations per training run). Instant-NGP moves most of that work into a multi-resolution hash table, shrinking the MLP to a tiny decoder.