Back to Academic/Research
Open WeightsSpecializedimage3dUpdated June 11, 2026

Surflo

Model Overview

Surflo is a undisclosed-parameter model developed by Academic/Research. Released on 2026-06-11.


📊 Quick Specs

Specification Table
SpecificationValue
Parametersundisclosed
Taskother
Modalityimage, 3d
LicenseOther/Custom
Typeopen-weights

✨ Key Features

  • Global Latent State Compression: Encodes arbitrary unposed views into a fixed-size 128-token latent using a Perceiver compressor over a VGGT-1B backbone
  • Arbitrary-Resolution Flow Matching Decoder: Decodes 3D points continuously from noise, enabling sampling from thousands to millions of surface points without spatial grid constraints
  • Inference-Time Photometric Guidance: Incorporates 3D Gaussian Splatting rendering loss gradients during ODE integration to resolve per-point decoding inconsistencies
  • Order-of-Magnitude Speedup: Provides fast feed-forward inference compared to compute-heavy optimization-based reconstruction algorithms
  • Unposed Multi-View Flexibility: Seamlessly handles between 2 and 64 unposed input images without pre-calibrated camera parameters

🔗 Resources


📜 License & Access

Other/Custom — See repository for specific license details.

Key Features

Global Latent State Compression: Encodes arbitrary unposed views into a fixed-size 128-token latent using a Perceiver compressor over a VGGT-1B backbone

Feature 01

Arbitrary-Resolution Flow Matching Decoder: Decodes 3D points continuously from noise, enabling sampling from thousands to millions of surface points without spatial grid constraints

Feature 02

Inference-Time Photometric Guidance: Incorporates 3D Gaussian Splatting rendering loss gradients during ODE integration to resolve per-point decoding inconsistencies

Feature 03

Order-of-Magnitude Speedup: Provides fast feed-forward inference compared to compute-heavy optimization-based reconstruction algorithms

Feature 04

Unposed Multi-View Flexibility: Seamlessly handles between 2 and 64 unposed input images without pre-calibrated camera parameters

Feature 05

You might also want to compare

Verified Sources

Tags

research-preview3d-reconstructionflow-matchingecole-polytechniquekyoto-universityuc-berkeley

Model Specs

open-weights

Parameters

Undisclosed

Context Window

undisclosed

License

Other/Custom

Deployment

self-hostable

Resources & Links

Curator Notes

Partially enriched via migration on 2026-07-25. Manual review recommended.

Compare Specs

Compare parameters, context windows, modalities, and benchmark scores of this model side-by-side with others.

Compare Model