Research map / Image generation / Image restoration
ColorFormer: Image Colorization via Color Memory Assisted Hybrid-Attention Transformer
Xiaozhong Ji, Boyuan Jiang, Donghao Luo, Guangpin Tao, Wenqing Chu, Zhifeng Xie, Chengjie Wang, Ying Tai
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace ResNet50 encoder with Swin-Transformer (ablation) | ColorFormer w/o CM · ImageNet val | FID: 4.10 -> 2.68 | Improved |
| replace Swin-Transformer with GLH-Transformer encoder (ablation) | ColorFormer w/o CM · ImageNet val | FID: 2.68 -> 1.85 | Improved |
| replace Twins with GLH-Transformer encoder (ablation) | ColorFormer w/o CM · ImageNet val | FID: 2.48 -> 1.85 | Improved |
| add Color Memory module (one group) (ablation) | ColorFormer GLH-Transformer · ImageNet val | FID: 1.85 -> 1.71 | Improved |
| add Color Memory module (four groups) (ablation) | ColorFormer GLH-Transformer · ImageNet val | CF: 36.70 -> 39.76 | Improved |
| use four groups of color priors instead of one (ablation) | ColorFormer · ImageNet val | CF: 36.96 -> 39.76 | Improved |
| initialize memory with Memory Build clustering instead of random (ablation) | ColorFormer CM4 · ImageNet val | FID: 1.80 -> 1.71 | Improved |
| increase memory size m from 256 to 1024 (ablation) | ColorFormer CM · ImageNet val | FID: 1.93 -> 1.68 | Improved |
| increase color embedding dimension d2 from 128 to 512 (ablation) | ColorFormer CM · ImageNet val | FID: 1.91 -> 1.65 | Improved |
| reduce semantic dimension k from 128 to 32 (ablation) | ColorFormer CM · ImageNet val | FID: 1.78 -> 1.87 | Hurt |
| use ColorFormer instead of Wu et al. (GAN-prior colorization) | automatic colorization · ImageNet val | FID: 3.62 -> 1.71 | Improved |
Later papers that compared with this method
2 later accepted papers report ColorFormer as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Automatic Controllable Colorization via Imagination (CVPR 2024 · better on 0 of 1 shared result)
- DDColor: Towards Photo-Realistic Image Colorization via Dual Decoders (ICCV 2023 · better on 5 of 5 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Colorization Transformer (ICLR 2021 · similarity 0.92)
- DDColor: Towards Photo-Realistic Image Colorization via Dual Decoders (ICCV 2023 · similarity 0.91)
- Learning Representations for Automatic Colorization (ECCV 2016 · similarity 0.90)
- Colorful Image Colorization (ECCV 2016 · similarity 0.90)
- CT2: Colorization Transformer via Color Tokens (ECCV 2022 · similarity 0.90)
- Pixelated Semantic Colorization (IJCV 2019 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
