facebookresearch/vggt

Health v2 · maintenance only

Flags: no_releases no_license

Adoption not part of the score

14292 stars · 1537 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-29, confidence not recorded

VGGT (Visual Geometry Grounded Transformer) is a feed-forward transformer model from Meta AI and Oxford VGG that infers 3D geometry—camera parameters, depth maps, and point maps—directly from one or more images. It ships as a Python library with pretrained checkpoints, a Hugging Face demo, and training/finetuning code.

Use cases

• estimate camera poses from multiple images
• reconstruct 3D point clouds from photos
• predict depth maps from a single or few images
• run structure-from-motion with a neural network instead of COLMAP
• finetune a 3D vision transformer on a custom dataset
• feed 3D geometry features into downstream vision or robotics models

When to choose

• you need fast feed-forward 3D reconstruction or camera estimation from images
• you want a learned alternative to classical SfM pipelines like COLMAP
• you need a strong pretrained 3D vision backbone for downstream tasks
• you want a CVPR 2025 award-winning, actively maintained research model with commercial-use checkpoint options

When to avoid

• you need a lightweight CPU-only solution—the model requires a GPU with substantial memory
• you need permissively licensed weights without an approval workflow—the original checkpoint is non-commercial and the commercial one requires an application form
• you need real-time on-device inference on edge hardware
• you need classical, highly precise bundle adjustment SfM with full control over the pipeline

Facets

library · maturity active

machine-learning deep-learning computer-vision image-processing transformers gpu-computing computer-vision machine-learning deep-learning artificial-intelligence robotics python cross-platform 3d-reconstruction visual-geometry transformer camera-estimation depth-estimation point-cloud cvpr-2025 research-model gpu linux

Member repositories

For agents

markdown · JSON · MCP: product_card(name="facebookresearch/vggt")