OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
Image & Video AI

Interactive Deep Colorization — richzhang

Interactive Deep Colorization — richzhang is a Python and PyTorch reimplementation for training and evaluating learned models that colorize images, including interactive colorization guided by user-provided color hints.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Interactive Deep Colorization — richzhang — OpenViST project profile in Image & Video AI.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use this repository for model training, evaluation, or obtaining weights for the related GUI workflow. Choose the separate interface repository when the immediate task is interactive, GUI-based colorization.

Who it is for

  • Researchers and developers training or evaluating image-colorization models in PyTorch.
  • Users who want to apply a trained model through the separate interactive GUI repository.

Documented uses

  • Prepare ILSVRC 2012 data for colorization experiments.
  • Train a model in automatic and interactive colorization stages.
  • Evaluate models on validation images or compare PSNR across different numbers of color hints.
  • Use a trained PyTorch model with the separate interactive GUI.
Before you choose

Requirements & limits

The repository is public and specifies an MIT license.

  • The repository centers on training and evaluation; its README points to a separate repository for the interactive GUI.
  • The listed prerequisites include PyTorch 0.4 or later, torchvision, and additional Python dependencies.
  • The documented training workflow uses ILSVRC 2012 data and has separate automatic and interactive fine-tuning stages.
  • The README distinguishes converted original weights from retrained weights, which it says behave qualitatively differently.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Prepare the dataset and train an interactive colorization model

Prepare ILSVRC 2012 data and run the repository's two-stage training workflow.

Before you start

  • Linux or macOS
  • Python 2 or 3
  • PyTorch 0.4 or later and torchvision
  • Repository dependencies, including examples Visdom and Dominate
  • ILSVRC 2012 dataset

Steps

  1. Install dependencies using the repository's requirements file and clone the repository.

  2. Prepare the dataset with the command below.

    Documented command
    python make_ilsvrc_dataset.py --in_path /PATH/TO/ILSVRC12

  3. Run ./scripts/train_siggraph.sh to train automatic colorization first and then fine-tune for interactive colorization.

Expected outcome

The README says the two stages produce checkpoints under `./checkpoints/siggraph_class` and `./checkpoints/siggraph_reg2`.

Guide limitations

  • The documented dataset-preparation process requires ILSVRC 2012.
  • The repository characterizes its implementation as an approximate reimplementation.
Documentation 1 ↗
Source-based practical guide

Test a model on validation images

Generate validation results for a pretrained or locally trained model.

Before you start

  • Installed repository dependencies
  • A pretrained model or a model produced by the training workflow

Steps

  1. Obtain a model by running pretrained_models/download_siggraph_model.sh or by training one in the repository.

  2. Run the documented test.py command for the selected model; for original Caffe weights, use --mask_cent 0.

  3. Open the generated HTML results under ./results/[[NAME]]/latest_val/index.html.

Expected outcome

The validation output includes automatic colorization and interactive colorization with a few or many randomly sampled hints.

Guide limitations

  • The README notes that the retrained model has better PSNR but behaves qualitatively differently from the converted original weights.
  • The documented tests use validation data prepared through the repository's dataset workflow.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
17 October 2026
Original website title
richzhang/colorization-pytorch · Official homepage title · retained in its source language
How we review sourcesSuggest a correction