Practical overview
When it fits
Use DDColor when you need model code and documented inference options for image colorization, or a starting point for integrating colorization into a larger application.
Who it is for
- Developers adding image colorization to a photo-editing or restoration workflow.
- Researchers and practitioners prototyping computer-vision colorization.
- People who want to run pretrained image-colorization inference locally.
Documented uses
- Add plausible color to black-and-white photographs, including historical images.
- Experiment with colorizing illustrated or game landscapes.
- Train a colorization model on ImageNet or a custom dataset.
- Export a supported model to ONNX.
Requirements & limits
The repository is public and lists the Apache-2.0 license. Local setup requires Python and PyTorch; a ModelScope inference route and online demo links are also documented.
- The repository documents local-script, Hugging Face, and ModelScope inference routes.
- The Model Zoo includes multiple pretrained model variants, including a tiny model.
- Training entails extra dependencies, dataset preparation, and separate ConvNeXt and InceptionV3 pretrained weights.
- ONNX export is documented with specific ONNX, ONNX Runtime, and onnxsim dependencies.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Run local colorization inference
Download a pretrained ModelScope checkpoint and pass an image directory to the local inference script.
Before you start
- Python 3.7 or later
- PyTorch 1.7 or later
- The DDColor repository and its requirements installed
- A pretrained DDColor checkpoint downloaded from ModelScope
Steps
Download the pretrained model with ModelScope's snapshot_download function, using the repository ID damo/cv_ddcolor_image-colorization.
Pass the downloaded pytorch_model.pt path and an input image directory to scripts/infer.py using the --model_path and --input options.
Expected outcome
The local inference script processes the supplied input directory with the pretrained DDColor model.
Guide limitations
- The example documents inference using a local script; it does not specify output handling or a user interface.
Prepare and start DDColor training
Set up the documented dataset and pretrained weights, configure training options, and start the training script.
Before you start
- Python 3.7 or later and PyTorch 1.7 or later
- ImageNet or a custom dataset
- Pretrained ConvNeXt and InceptionV3 weights
- The additional training requirements and basicsr
Steps
Prepare an ImageNet or custom dataset and use scripts/get_meta_file.py to create the dataset list file.
Place the pretrained ConvNeXt and InceptionV3 weights in the pretrain folder.
Set meta_info_file and other options in options/train/train_ddcolor.yml.
Start training with sh scripts/train.sh.
Expected outcome
The documented training procedure is configured and launched using the selected dataset and pretrained weights.
Guide limitations
- The README does not specify training duration, hardware requirements, or expected model quality.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 14 October 2026
- Original website title
- piddnad/DDColor · Official homepage title · retained in its source language