test213 / README.md
dkescape's picture
Upload 10 files
0e868b4 verified

A newer version of the Gradio SDK is available: 6.9.0

Upgrade
metadata
title: Color Restorization Model
emoji: 🖼️
colorFrom: indigo
colorTo: yellow
sdk: gradio
sdk_version: 3.9
app_file: app.py
pinned: true
license: unlicense

🌈 Color Restorization Model (CPU Optimized)

Bring your old black & white photos back to life—upload, adjust, and download in vivid color.

This version has been optimized for CPU inference, removing GPU dependencies and improving performance on standard hardware.

Features

  • Adaptive Resolution Processing: Large images are processed intelligently to preserve sharpness while ensuring fast colorization.
  • Quality Presets: Choose between Fast, Balanced, and High quality to suit your hardware.
  • Real-time Progress: Visual progress bar.
  • Pure CPU Stack: Optimized for Intel/AMD CPUs with AVX2 support (via PyTorch).

CPU Compatibility Matrix

Processor Generation Recommended Preset 1080p Processing Time (Est.)
Intel Core i3 / Older Fast (256px) 2-5s
Intel Core i5 (8th Gen+) Balanced (512px) 1-3s
Intel Core i7 / Ryzen 7 High (1080px) 3-8s
M1/M2 Mac Balanced <1s

Performance Tuning

  • Memory Constrained (<8GB RAM): Stick to "Fast" or "Balanced".
  • High-Res Archival: Use "Original" resolution only if you have >16GB RAM and patience.
  • Batch Processing: The core logic is thread-safe and can be extended for batch processing.

Technical Details

The application uses the DDColor architecture via ModelScope. Optimizations include:

  1. L-Channel Preservation: We apply colorization at a lower resolution and merge it with the original high-resolution Luminance channel using LAB color space.
  2. In-Memory Pipeline: Removed disk I/O bottlenecks.
  3. Dynamic Quantization: Automatically applied to the model on supported CPUs.

Installation

pip install -r requirements.txt
python app.py

Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference