DDColor
DDColor is an open-source PyTorch implementation of the ICCV 2023 paper Towards Photo-Realistic Image Colorization via Dual Decoders, developed by researchers at DAMO Academy, Alibaba Group. It represents a state-of-the-art approach to automatic image colorization, utilizing a dual-decoder architecture that optimizes learnable color tokens based on multi-scale visual features to generate vivid and natural results. The software is designed to restore historical black-and-white photographs and can also recolor animated scenes from games or anime to achieve realistic, real-life aesthetics. DDColor offers high flexibility with multiple pre-trained models available, including a lightweight tiny variant, and supports various deployment methods. Users can run inference locally using Python scripts, access the technology through online demos on ModelScope and Replicate, or integrate it directly into Python workflows via the Hugging Face Hub. The project requires Python 3.7 or higher and PyTorch 1.7 or higher. Install