# Xharlie/pointnerf

Point-NeRF: Point-based Neural Radiance Fields

Repository: https://github.com/Xharlie/pointnerf
Canonical: https://ross.abutalabs.com/products/pointnerf
Language: Python
License: NOASSERTION
License Family: other
Topics: nerf, point-cloud, point-based-graphics, volume-rendering, differentiable-rendering, neural-rendering, neural-renderer, mvs, multiview-stereo, reconstruction, 3d, graphics, 3d-reconstruction, scene-reconstruction, radiance-field, neural-radiance-fields, implicit-functions, tanks-and-temples, scannet, cvpr2022
Last push: 2024-03-08T20:57:52+00:00

## Health v2 (maintenance only)
Score: 32/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 0, release rhythm 35, longevity 100
- inputs: {"age_days": 1664, "days_push": 908, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, no_license
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1154, forks 128 (observed 2026-08-28T04:03:47.532231+00:00)

## What it is
Point-NeRF is a research implementation of a point-based neural radiance field method (CVPR 2022 Oral) that models scenes with neural 3D point clouds and renders them via ray marching. It can be initialized from a pre-trained deep network (e.g., MVSNet) and finetuned to surpass NeRF quality with roughly 30X faster training.

## Use cases
- reconstruct 3D scenes from multi-view images as neural radiance fields
- render novel views of a scene from a point-based radiance field
- initialize a NeRF from MVSNet point clouds and finetune it quickly
- benchmark novel view synthesis on NeRF-Synthetic, ScanNet, and Tanks and Temples
- research point-based neural rendering and point pruning/growing mechanisms

## When to choose
- you need a fast-training NeRF variant initialized from multi-view stereo point clouds
- you are researching point-based radiance fields or differentiable volume rendering
- you want a published CVPR 2022 baseline for novel view synthesis

## When to avoid
- you need a production-ready 3D reconstruction pipeline with support and licensing guarantees
- you want a simple off-the-shelf NeRF without GPU/CUDA setup
- you need a permissively licensed project (license is non-standard)

## Facets
- artifact type: library
- maturity: maintenance
- function: machine-learning, deep-learning, computer-vision, image-processing, graphics, simulation
- domain: computer-vision, graphics, machine-learning, deep-learning
- platform: python
- tags: nerf, neural-radiance-fields, point-cloud, volume-rendering, differentiable-rendering, neural-rendering, multiview-stereo, 3d-reconstruction, cvpr-2022, research-code, linux, gpu

## Member repositories
- Xharlie/pointnerf (main) score 32

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:03:47.532231+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T06:32:16.082089+00:00, confidence not recorded.
  - readme: https://github.com/Xharlie/pointnerf (fetched 2026-08-28T04:03:47.532231+00:00, sha 85cf8d81ef47)
- Data as of 2026-08-30T08:39:29.467469+00:00.