Skip to content

Latest commit

 

History

History
52 lines (42 loc) · 2.66 KB

README.md

File metadata and controls

52 lines (42 loc) · 2.66 KB

Colorization

Towards More Vibrant Colorful Image Colorization

Final project for CSE 203B Convex Optimization Course

Sample output images: (The groundtruth of #1490 is grayscale. gamma is color_vivid_gamma in our loss function. retrain means retraining for 10 epochs; otherwise, it's training for 20 epochs.)

Group Members

  • Ahan Mukhopadhyay
  • Kolin Guo
  • Kyle Lisenbee
  • Ulyana Tkachenko

Prerequisites

  • Ubuntu 18.04
  • NVIDIA GPU with CUDA version >= 11.1, cuDNN version >= 8.0
  • Docker version >= 19.03, API >= 1.40
  • nvidia-container-toolkit (previously known as nvidia-docker)

Command to test if all prerequisites are met:
sudo docker run -it --rm --gpus all ubuntu nvidia-smi

Setup Instructions

bash ./setup.sh
You should be greeted by the Docker container colorization when this script finishes. The working directory is / and the repo is mounted at /Colorization.

Running Instructions

  • Training from scratch
    python3 src/train.py eccv16_half
    Resume training from a checkpoint file
    python3 src/train.py eccv16_half --checkpoint_file <path_to_ckpt_file>
  • Predicting: generate colorized images using trained models
    python3 src/predict.py eccv16_half --checkpoint_file <path_to_ckpt_file>

Some other available arguments can be viewed with --help option.

Presentation and Report

Our project outline and final report can be found in docs/ folder.

Additional Work

  • Change the network to output chroma/hue channels instead of ab channels and use a similar loss function to constrain on chroma/hue directly.
  • Benchmark against the multinomial classification loss function in [1].

Credit

  1. R. Zhang, P. Isola, and A. A. Efros, "Colorful Image Colorization," in ECCV, 2016.
  2. R. Zhang, J. Zhu, P. Isola, X. Geng, A. S. Lin, T. Yu, A. A. Efros, " Real-Time User-Guided Image Colorization with Learned Deep Priors," in SIGGRAPH ,2017.
  3. GitHub of [1], [2]
  4. GitHub of [2] in PyTorch