Computer Vision
Prizm AI
U-Net Semantic Segmentation Pipeline
Implemented advanced semantic segmentation with ResNet/VGG19 backbones achieving 98% pixel accuracy for vector-based design workflows. Used Label-Studio for annotation, Albumentations for augmentation, and developed RGB-to-SVG conversion pipeline processing 10k+ images with white rendering optimization.
3 months
Project Duration
98% pixel accuracy
Key Achievement
6+
Technologies Used
The Challenge
Prizm AI needed a highly accurate semantic segmentation pipeline to convert raster images into vector-based designs for their creative workflow platform.
The Solution
Implemented a sophisticated U-Net architecture with ResNet and VGG19 backbones, integrated with Label Studio for efficient data annotation and preprocessing pipeline handling 10k+ images.
Results & Impact
- Achieved 98% pixel-level accuracy in semantic segmentation
- Processed over 10,000 images with automated preprocessing
- Reduced manual design conversion time by 85%
- Enabled real-time RGB-to-SVG conversion capabilities
Project Metrics
98% pixel accuracy
10k+ image preprocessing
Technologies Used
U-NetResNetVGG19Label StudioOpenCVRGB-to-SVG
Project Details
Company
Prizm AI
Duration
3 months
Team Size
Solo project with design team collaboration
Category
Computer Vision