Skip to content

I want to create a synthetic data set to get a feeling on how well modern vision models can recognize realworld images based on synthetic data.

Notifications You must be signed in to change notification settings

BjoernGit/PerceptionAutoLabeling

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

16 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

PerceptionAutoLabeling

Overview

This project explores the effectiveness of modern computer vision models when trained exclusively on synthetic datasets generated using Unity's Perception package.

Methodology

The foundation of this project follows Unity's comprehensive tutorial for synthetic data generation: https://docs.unity3d.com/Packages/com.unity.perception@1.0/manual/Tutorial/Phase1.html

Custom 3D Asset Creation

To create a unique test case, I 3D scanned a wooden cobra sculpture using a high-resolution Creality Raptor 3D scanner, capturing intricate details for realistic synthetic training data.

3D Scanned Wooden Cobra Sculpture

Dataset Generation

The scanned cobra model was integrated into Unity scenes to automatically generate a comprehensive dataset of 10,000 synthetic images with varied lighting, backgrounds, and camera angles.

Training Process

Training Demo

🎬 Full Training Video (MP4)

Results

https://github.com/BjoernGit/Unity-perception-to-yolo

About

I want to create a synthetic data set to get a feeling on how well modern vision models can recognize realworld images based on synthetic data.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published