# yyfz/Pi3

[ICLR 2026] π^3: Permutation-Equivariant Visual Geometry Learning

Repository: https://github.com/yyfz/Pi3
Canonical: https://ross.abutalabs.com/products/pi3
Homepage: https://yyfz.github.io/pi3/
Language: Python
License: BSD-3-Clause
License Family: permissive
Last push: 2026-07-03T13:17:10+00:00

## Health v2 (maintenance only)
Score: 59/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 90, release rhythm 35, longevity 29
- inputs: {"age_days": 413, "days_push": 61, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 2122, forks 167 (observed 2026-08-28T04:06:16.615653+00:00)

## What it is
Pi3 (π³) is a feed-forward neural network for visual geometry reconstruction that eliminates the need for a fixed reference view, using a permutation-equivariant architecture to achieve state-of-the-art results. It reconstructs point clouds, camera poses, and depth from images or videos, with an upgraded Pi3X model supporting conditional injection and approximate metric scale.

## Use cases
- reconstruct 3d point clouds from photos
- estimate camera poses from a video
- predict depth maps from images without a reference view
- recover visual geometry from in-the-wild videos
- run 3d reconstruction inference on gpu
- train a visual geometry model
- evaluate multi-view geometry reconstruction models

## When to choose
- you need feed-forward 3d reconstruction from multiple images or video without fixed reference view
- you want state-of-the-art point cloud, camera pose, and depth prediction in one model
- you need conditional injection of camera pose, intrinsics, or depth (Pi3X)
- you want a research-grade model with training, evaluation, and demo code available

## When to avoid
- you need real-time reconstruction on CPU or edge devices
- you need a lightweight classical SfM/SLAM pipeline like COLMAP
- you cannot run GPU inference
- you need production-supported software with long-term maintenance guarantees

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, deep-learning, computer-vision, image-processing, graphics, simulation
- domain: computer-vision, deep-learning, machine-learning, graphics, artificial-intelligence
- platform: python, cross-platform
- tags: 3d-reconstruction, visual-geometry, point-cloud, camera-pose-estimation, depth-estimation, feed-forward-network, permutation-equivariant, structure-from-motion, research-code, iclr-2026, gpu, linux

## Member repositories
- yyfz/Pi3 (main) score 59

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:06:16.615653+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 2026-08-30T02:52:52.354140+00:00, confidence not recorded.
  - readme: https://github.com/yyfz/Pi3 (fetched 2026-08-28T04:06:16.615653+00:00, sha e7498f8f2245)
  - homepage: https://yyfz.github.io/pi3/ (fetched 2026-08-29T10:32:49.950144+00:00, sha 5f77c59fd48b)
- Data as of 2026-08-30T08:39:29.467469+00:00.