vLAR-group

vLAR-group / PhysInOne

Public

🔥PhysInOne in Python (CVPR 2026)

32
0
100% credibility
Found Apr 14, 2026 at 32 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
AI Summary

PhysInOne is a massive collection of simulated physics videos with detailed labels designed to help AI learn about real-world physical phenomena.

How It Works

1
🔍 Discover PhysInOne

You stumble upon PhysInOne while looking for fun ways to learn about physics through videos.

2
📖 Read the Overview

You check out the main page and get excited by the huge collection of over 2 million videos showing everyday physics like bouncing balls, flowing water, and magnets in action.

3
🎥 Explore the Teaser

You watch the colorful teaser video that showcases realistic scenes from different angles, sparking your curiosity about real-world physics.

4
🌐 Visit Project Links

You click to the project page or paper to learn more about the 71 physics topics and thousands of objects in natural settings.

5
📥 Get the Videos

You head to the data sharing site and download the massive set of videos with helpful labels on movements, materials, and descriptions.

✅ Dive into Physics

Now you can watch, study, and use these videos to understand or teach physics in a visual, engaging way that feels just like real life.

Sign up to see the full architecture

4 more

Sign Up Free

Star Growth

See how this repo grew from 32 to 32 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is PhysInOne?

PhysInOne is a massive Python-based dataset for visual physics learning, packing 2 million videos from 153,810 dynamic 3D scenes that cover 71 physical phenomena across mechanics, optics, fluid dynamics, and magnetism. It solves the shortage of grounded physics data for training AI models by providing rich annotations like 3D geometry, object dynamics, physical properties, and natural-language descriptions, all accessible via Hugging Face. Developers get a one-stop suite for tasks like physics-aware video generation and future frame prediction.

Why is it gaining traction?

This CVPR 2026 submission stands out with its unprecedented scale—2M videos featuring multi-object interactions in realistic environments with 2,231 objects and 623 materials—far beyond smaller physics datasets. The hook is immediate access to the full dataset on Hugging Face, plus a detailed arXiv paper and project page, letting researchers benchmark models without waiting for code. Early adopters appreciate the multi-view captures (13 per scene) for robust training.

Who should use this?

Computer vision researchers building physics-informed models for video prediction or property estimation. AI engineers at robotics firms simulating real-world interactions with fluids or magnets. ML teams in simulation software needing diverse, annotated scenes for generative tasks.

Verdict

Promising dataset for physics AI, but skip for now—1.0% credibility score, just 26 stars, and no code yet despite "coming soon" promises. Check the Hugging Face dataset and paper first; revisit post-release for production use.

(178 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.