Research map / Detection and segmentation / CNN features
Pixelated Semantic Colorization
Jiaojiao Zhao, Jungong Han, Ling Shao, Cees G. M. Snoek
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 |
|---|---|---|---|
| add pixelated semantic embedding and generator guidance | pixelated colorization PixelCNN · Pascal VOC2012 val | segmentation mean-IoU of colorized images: 67.6 -> 70.3 vs strongest-listed baselines; ours 70.3 vs 68.8 | Improved |
| add semantic segmentation branch guiding the color embedding (ablation) | pixelated colorization · Pascal VOC2012 and COCO-stuff | visual quality | Improved |
| add semantic labels to condition the generator (concatenation) (ablation) | pixelated semantic colorization · Pascal VOC2012 and COCO-stuff | visual quality | Improved |
| use channel concatenation instead of feature transformation to inject segmentation into the generator (ablation) | pixelated semantic generator · Pascal VOC2012 | visual quality | Improved |
| colorize gray images before segmentation | Deeplab-ResNet101 · Pascal VOC2012 val | mean-IoU: 66.9 -> 70.3 | Improved |
| use original color instead of our colorized image | Deeplab-ResNet101 · Pascal VOC2012 val | mean-IoU: 70.3 -> 72.1 | Improved |
| initialize segmentation from pre-trained colorization instead of scratch (ablation) | segmentation branch · Pascal VOC2012 | validation loss: 0.060 -> 0.043 | Improved |
| jointly train segmentation and colorization instead of pre-trained colorization only (ablation) | segmentation branch · Pascal VOC2012 | validation loss: 0.043 -> 0.030 | Improved |
| compare PSNR of our semantic method with best baseline | pixelated semantic colorization · Pascal VOC2012 val | PSNR: 24.56 -> 23.15 | Hurt |
| add pixelated semantics to the model (ablation) | pixelated colorization · Pascal VOC2012 val | sample diversity (MS-SSIM) | No clear effect |
| use ours versus best diverse baseline for naturalness | pixelated semantic colorization · VOC + COCO-stuff | naturalness (%): 89.89 -> 94.65 | Improved |
Closest papers by meaning
- Learning Representations for Automatic Colorization (ECCV 2016 · similarity 0.92)
- Colorful Image Colorization (ECCV 2016 · similarity 0.92)
- Instance-Aware Image Colorization (CVPR 2020 · similarity 0.91)
- DDColor: Towards Photo-Realistic Image Colorization via Dual Decoders (ICCV 2023 · similarity 0.90)
- ColorConceptBench: A Benchmark for Probabilistic Color-Concept Understanding in Text-to-Image Models (NeurIPS 2026 · similarity 0.88)
- Automatic Controllable Colorization via Imagination (CVPR 2024 · similarity 0.88)
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.
