pi3 / README.md
Flor's picture
card: update from fastforks 1cbf1671
a9a4ae0 verified
|
Raw History Blame Contribute Delete
6.55 kB
metadata
library_name: fastforks
pipeline_tag: image-to-3d
license: cc-by-nc-4.0
tags:
  - multiview
  - multi-view-stereo
  - 3d-reconstruction
  - camera-pose-estimation
  - cuda
  - pytorch
  - fastforks
  - fastforks-card

Pi3 (fastforks)

π³, permutation-equivariant visual geometry learning (Wang et al. 2025): cameras, depth and point maps of every view from unposed images in one forward pass, with no reference view. A DINOv2 ViT-L/14 encoder feeds 36 alternating frame / global attention blocks (q/k norm, 2-D RoPE); three small transformer decoders give local point maps, confidences and camera poses.

This repo is the model card only: it holds no weights. fastforks downloads the authors' weights from the links below and checks them against the SHA-256 in the model card. Code: https://github.com/kleinicke/fastforks. Kernels for the kernels backend: https://huggingface.co/fastforks/kernels. Original code: https://github.com/yyfz/Pi3 at 9fa3ddb.

Author weight repos: https://huggingface.co/yyfz233/Pi3.

Task: multiview. Tier: experimental. Outputs: depth, confidence, world_points, cam_to_world. Backends: portable, kernels.

Input: several images of one scene (posed and calibrated for multi-view stereo, unposed otherwise). Output: depth maps and, where the model predicts them, cameras and 3-D points.

Variants

Weights are the authors' files (not hosted here). SHA-256 shortened to 12 hex digits; the full hashes are in card.toml.

variant weights (author link) SHA-256 weights licence commercial use
pi3@1 model.safetensors 33580e4702ac CC-BY-NC-4.0 no

Licence

Original-code strings, recorded as they are (research-use package; check them before any other use). Code: BSD-3-Clause. Licence files: LICENSE.

Training data: Per the paper (arXiv 2507.13347): a mixture of public synthetic and real multi-view datasets; DINOv2 initialisation. The README restricts the weights to non-commercial use because of the training datasets..

  • Licence note: The Pi3 code is BSD-3-Clause, but the README licenses the Pi3 weights CC-BY-NC-4.0 (non-commercial); the Hugging Face card's bsd-2-clause tag is overridden by that statement.

Licence evidence: https://github.com/yyfz/Pi3/blob/9fa3ddb/LICENSE and https://github.com/yyfz/Pi3/blob/9fa3ddb/README.md#-license

Input size

any H x W >= 1, resized or padded inside the model Tested up to (portable, fp16, RTX 5060 Ti 16 GB): pi3 3000x4000 (4.8 GiB peak allocated, 6.5 reserved)

Speed and memory

Multiview, synthetic

model input time (ms): original fp32 original portable kernels kernels speed-up: vs original vs portable peak memory (MiB, allocated / reserved): original portable kernels
pi3 336x518 – 1057 821 560 1.9x 1.5x 5926 / 6016 2360 / 2416 2367 / 2456

Startup and first call

model load (s): original portable kernels first call (s): original portable kernels
pi3 10.64 (336x518) 2.35 (336x518) 3.54 (336x518) 4.06 (336x518) 1.24 (336x518) 0.69 (336x518)

torch.compile and TensorRT

model input time: portable compile TensorRT fp16 kernels preparation: compile (cold first call) ONNX + engine build peak MiB (alloc / reserved): portable compile kernels TensorRT (engine + torch)
pi3 336x518 821 ms 752 ms planned for 0.2.1 560 ms 55.6 s – 2360 / 2416 2205 / 2334 2367 / 2456 –

Fixed-size cells: time (ms) / peak allocated / peak reserved memory (MiB).

Hardware, protocol and caveats: benchmarks (RTX 5060 Ti 16 GB, sm_120). Original = the original code named above at the precision of portable (fp16, or (fp32) where fp16 is unsupported or not finite); original fp32 = its default fp32 run. Peak memory is the torch allocator's.

Accuracy

On public datasets

640x512 crops, the 7 most co-visible views as sources; GT = depth rendered from the structured-light points. AbsRel / delta1 over GT pixels, after per-image median scaling unless the model takes the cameras (calibrated MVS: cameras, error in mm; PatchMatchNet also the depth range 425-935 mm, MVSAnywhere its own range from the camera baselines). Original = the original code named under Details, same checkpoint and protocol.

variant dataset original fp32 original fp16 portable kernels note
pi3 DTU scans 1, 9, 24, 33, 75, view 24 + 7 sources, AbsRel / delta1 % 0.0070 / 99.94 0.0071 / 99.94 0.0070 / 99.94 0.0070 / 99.94

Measured on NVIDIA GeForce RTX 5060 Ti, torch 2.14.1+cu130, fastforks 28db82cb, 2026-11-13.

Parity with the original code

Every variant is checked against reference outputs of the original code at its pinned commit. CPU, float32: max |diff| <= 0.002 vs the original. GPU, float16 (the default), portable and kernels: mean |diff| <= 0.02 and p99 <= 0.25 (absolute error of the output depth).

Usage

from fastforks import multiview

# non-commercial weights: run with FASTFORKS_NONCOMMERCIAL=1 in the environment
model = multiview.load("pi3")  # mode="auto": the preferred backend from recorded measurements
result = model(["view0.jpg", "view1.jpg", "view2.jpg"])  # the views of one scene, no cameras needed
result.depth  # [V, H, W] float32 tensor; result.kind says what the maps mean
result.backend  # what ran, e.g. Backend(mode='portable', device='cuda:0', dtype='float16')

Install with pip install fastforks (see https://github.com/kleinicke/fastforks). Other variants: pi3.

Kernels backend

multiview.load("pi3", mode="kernels") runs our CUDA kernels (fastbb). It needs a prebuilt kernel for your GPU architecture (downloaded from https://huggingface.co/fastforks/kernels on first use) and a recent NVIDIA driver; if there is none it raises BackendUnavailableError with the reason and never falls back silently. mode="auto" picks kernels when it can run and logs the choice.