colorization and interactive-deep-colorization
These are complementary tools where the second builds upon the first's neural network architecture—junyanz/interactive-deep-colorization extends richzhang/colorization's automatic approach by adding interactive user guidance, allowing users to refine colorization results with manual color hints rather than relying solely on the model's predictions.
About colorization
richzhang/colorization
Automatic colorization using deep neural networks. "Colorful Image Colorization." In ECCV, 2016.
Uses a CNN trained on Lab color space to predict chrominance from grayscale luminance, with PyTorch inference optimized for minimal dependencies. Includes both fully automatic colorization and an interactive variant enabling user-guided hints via the SIGGRAPH 2017 model. Provides pretrained weights and preprocessing/postprocessing pipelines (Lab conversion, 256×256 resizing, full-resolution reconstruction) for straightforward integration into Python workflows.
About interactive-deep-colorization
junyanz/interactive-deep-colorization
Deep learning software for colorizing black and white images with a few clicks.
Implements a Local Hints Network that enables real-time interactive colorization through user-placed color hints on specific image regions, which the model propagates intelligently across the image. Supports both Caffe (official SIGGRAPH 2017 model) and PyTorch backends, with a full Qt5-based GUI for point placement, color selection, and patch-size adjustment, alongside Jupyter notebook interfaces for lighter-weight integration. Built on learned deep priors that generalize color propagation from sparse user input, also including a Global Hints Network variant for histogram-based color transfer applications.
Related comparisons
Scores updated daily from GitHub, PyPI, and npm data. How scores work