LongSplat

Visit Tool

LongSplat is a Content & Design tool that offers robust unposed 3D Gaussian Splatting for casual long videos. It features incremental joint optimization for high-quality novel view synthesis while remaining memory-efficient.

Claim this tool

1View

At a glance

Pricing

Open Source

Free tier

Yes

API

Skill level

Technical

About

What is LongSplat?

LongSplat is an advanced framework designed for robust 3D Gaussian Splatting from casually captured long videos. It addresses the challenges of unposed video data through incremental joint optimization, which includes pose estimation with MASt3R and adaptive octree anchoring. This allows for high-quality novel view synthesis from free viewpoints, making it suitable for various applications in computer vision and 3D graphics. The tool is memory-efficient and scalable, capable of handling varied resolutions and frame rates. It also provides a conversion script to transform its results into the standard 3DGS format for compatibility with general 3DGS viewers.

Best used for

Ideal for developers and data scientists who need to reconstruct robust 3D scenes from casual long videos, generate high-quality novel views, and convert results to standard 3DGS format. Especially valuable for research and development in computer vision and 3D graphics.

Common actions

reconstruct 3D scenes

generate novel views

process video data

optimize 3D models

github copilotface swapping"AI Agents"workflowsopen-sourcecollaborationlow-code/no-codedeepfakeautomated workflow

Capabilities

Key features

Unposed 3D Gaussian Splatting
Incremental joint optimization
Pose estimation with MASt3R
Adaptive octree anchoring
3DGS format conversion
Memory-efficient reconstruction

Target Audience

developerdata scientist

Integrations

Not yet documented

Pricing & Plans

Open Source

Free

FAQs

What kind of videos can LongSplat process effectively?

LongSplat is designed for casually captured long videos, handling varied resolutions and frame rates. For optimal results, it's suggested to subsample videos to around 10 fps and resize to 512 px width. It performs best with videos lacking dynamic objects and having sufficient camera overlap.

How does LongSplat handle unposed video data?

LongSplat addresses unposed video data through incremental joint optimization. This process integrates pose estimation using MASt3R and adaptive octree anchoring, allowing the framework to robustly reconstruct 3D scenes and synthesize novel views even without pre-defined camera poses.

Can LongSplat's output be used with other 3D Gaussian Splatting viewers?

Yes, LongSplat provides a conversion script to transform its results into the standard 3DGS format. This generates a point_cloud.ply file, which ensures compatibility with general 3DGS viewers like Three.js. Pre-pruning is recommended before conversion to optimize model size.

Trending

Subcategories trending in Content & Design

Image Generation AI Writing Assistants Audio & Music Video Generation Photo Editing Graphic Design

Trending

Explore

Browse AI tools by category

Content & Design Productivity & Business Coding & Development AI Agents & Automation Research & Education Wellness & Lifestyle Career Development Marketing & Growth Data & Analytics Customer Support & CX Finance E-commerce