NeRF Service backend: runpod

A small service that turns a set of overlapping photos into a 3D scene. You upload frames, it runs structure-from-motion + Gaussian-splatting on a GPU worker, and returns a .ply. The job takes minutes, so the API is asynchronous: submit → poll → download (HTTP 202 + polling).

Try it

• Open the interactive API docs (/docs) — drive the whole flow from your browser: /upload a few overlapping photos, /nerfify, poll /jobs/<id>, then /jobs/<id>/result.
• Download a sample result (a real splat from 20 photos) — then drag the .ply into a viewer like antimatter15.com/splat or SuperSplat
• Source (FastAPI API + RunPod GPU worker + client): github.com/mtbouchard/nerf-service
• Health check (/healthz)

API

POST /upload          multipart "file"          -> {"id": "..."}   (repeat per frame)
POST /nerfify         {"images": [id1, ...]}    -> 202 {"job_id": "...", "status": "pending"}
GET  /jobs/<job_id>                              -> {"status": "running", ...}
GET  /jobs/<job_id>/result                       -> the .ply file