MakeItColor: Image Colorization Model

Overview

MakeItColor is a deep learning model designed for automatic image colorization. It transforms grayscale images into vivid, realistic colorized outputs using a PyTorch-based Convolutional Neural Network (CNN) architecture integrated with the ModelScope framework.

This model builds upon the work of DDColor, utilizing a dual-encoder approach and trained on the ImageNet-Val5k dataset.

Features

Task: Image Colorization
Framework: PyTorch, ModelScope
Architecture: Convolutional Neural Network (CNN)
Input: Grayscale images (single-channel)
Output: Colorized images (RGB format)

Installation

Ensure you have Python 3.7+ installed. Then, install the required dependencies:

pip install opencv-python
pip install modelscope==1.12.0 
pip install datasets==2.14.7
pip install pillow
pip install numpy

Usage

ModelScope Pipeline

import cv2
from modelscope.pipelines import pipeline
from modelscope.utils.constant import Tasks
from huggingface_hub import snapshot_download

# Download the model files to a local directory
snapshot_download(repo_id="muhammadnoman76/makeitcolor", local_dir="./makeitcolor", repo_type="model")

# Initialize the colorization pipeline
img_colorization = pipeline(Tasks.image_colorization, model='./makeitcolor')

# Load a grayscale image
img_path = 'input.jpg'

# Run colorization
result = img_colorization(img_path)

# Save the colorized image
cv2.imwrite('result.png', result['output_img'])

Note:

Ensure that the input image (input.jpg) is a proper grayscale (single-channel) image.

The output (result.png) will be a standard RGB image.

Google Colab

For an interactive demonstration, try our Google Colab notebook.

Model Files

The repository contains:

pytorch_model.pt: Pre-trained model weights
configuration.json: Model configuration file for ModelScope integration
README.md: Documentation

Requirements

Hardware

CPU (supported)
GPU (recommended for faster inference)

Software Dependencies

modelscope
opencv-python
torch

Input Format

Grayscale images (.png, .jpg, etc.)

Example Workflow

Prepare a grayscale image (e.g., input.jpg)
Run the provided example code
Check the output file (result.png) for the colorized result

Limitations

May struggle with highly complex, ambiguous, or abstract grayscale images
Performance and output quality depend on the clarity and details of the input
Primarily optimized for natural images; results may vary for synthetic or artistic inputs

Credits

This work builds upon and was inspired by the DDColor project.
MakeItColor leverages a dual-encoder strategy from DDColor and is trained on the ImageNet-Val5k dataset.

Special thanks to the creators of DDColor for their foundational contributions.

License

This project is licensed under the Apache License 2.0.

Contact

For issues, questions, or feedback:

Open an issue on the Hugging Face repository
Contact the maintainer directly at: [email protected]

Developed by Muhammad Noman