Research map / Image generation / Editing and style transfer
Structural Consistency and Controllability for Diverse Colorization
Safa Messaoud, David Forsyth, Alexander G. Schwing
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 a fully connected G-CRF output layer to the VAE-MDN colorizer | VAE-MDN · LFW | error-of-best per pixel: .035 -> 11 e^{-5} | Improved |
| add a fully connected G-CRF output layer to the VAE-MDN colorizer | VAE-MDN · LSUN-Church | error-of-best per pixel: .028 -> 93 e^{-6} | Improved |
| add a fully connected G-CRF output layer to the VAE-MDN colorizer | VAE-MDN · ILSVRC-2015 | error-of-best per pixel: .033 -> 12e^{-5} | Improved |
| add a G-CRF structured output layer to VAE-MDN | VAE-MDN · LFW user study | % preferring Ours vs VAE-MDN: 61.12 % | Improved |
| add a G-CRF structured output layer to VAE-MDN | VAE-MDN · LSUN-Church user study | % preferring Ours vs VAE-MDN: 66.89 % | Improved |
| add a G-CRF structured output layer to VAE-MDN | VAE-MDN · ILSVRC-2015 user study | % preferring Ours vs VAE-MDN: 54.79 % | Improved |
| add a G-CRF layer to VAE-MDN (diversity measured by pairwise SSIM) | VAE-MDN · LFW, LSUN-Church, ILSVRC-2015 | mean pairwise SSIM (lower = more diverse): .49 -> .61 (LFW), .77 -> .83 (LSUN-Church), .48 -> .52 (ILSVRC-2015) | Hurt |
| use the learned attention-like G-CRF structure instead of local smoothness for propagating sparse edits | Levin et al. propagation · ImageNet test, 100 edits | PSNR (dB): 30 -> 30.4 | Improved |
| add G-CRF output layer to the feature-consistent VAE (S-VAE) | Feature Consistent VAE · TFD | KL divergence: 0.16 -> 0.13 | Improved |
| add G-CRF layer on the BEGAN discriminator (S-BEGAN) | BEGAN · TFD | KL divergence: 0.09 -> 0.07 | Improved |
| add G-CRF layer to VAE for sharpness (Laplacian variance) | Feature Consistent VAE · TFD | Laplacian sharpness: 30.17% -> 92.% | Improved |
Closest papers by meaning
- Learning Diverse Image Colorization (CVPR 2017 · similarity 0.91)
- Beyond Fixed Luminance: Towards Panchromatic and Orthochromatic Image Colorization (ECCV 2026 · similarity 0.88)
- Automatic Controllable Colorization via Imagination (CVPR 2024 · similarity 0.88)
- Recolour What Matters: Region-Aware Colour Editing via Token-Level Diffusion (ECCV 2026 · similarity 0.88)
- Control Color: Multimodal Diffusion-Based Interactive Image Colorization (IJCV 2025 · similarity 0.88)
- Enabling Region-Specific Control via Lassos in Point-Based Colorization (AAAI 2025 · similarity 0.86)
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.
