Research map / Image generation / Image restoration
Deep Image Harmonization with Learnable Augmentation
Li Niu, Junyan Cao, Wenyan Cong, Liqing Zhang
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 reconstruction branch (bijection between latent codes and original composites) (ablation) | SycoNet · Hday2night | Div: 0.01 -> 1132.71 | Improved |
| add reconstruction branch (ablation) | SycoNet · Hday2night | fMSE: 569.49 -> 517.00 | Improved |
| feed both real and composite images to latent encoder E^z (ablation) | SycoNet · Hday2night | fMSE: 554.48 -> 517.00 | Improved |
| use softmax-normalized coefficients and clustered LUT initialization instead of prior LUT design (ablation) | SycoNet · Hday2night | fMSE: 541.56 -> 517.00 | Improved |
| append random vector z at E^z input rather than inside (ablation) | SycoNet · Hday2night | fMSE: 517.00 -> 525.54 | Hurt |
| sample z^c (reconstruct originals) instead of random z^g (ablation) | SycoNet · Hday2night | fMSE: 517.00 -> 556.44 | Hurt |
| augment with traditional color transfer instead of SycoNet | iS2AM finetuning on Hday2night/HFlickr · iHarmony4 small datasets | MSE, fMSE, fSSIM | Hurt |
| augment with random internet LUTs | iS2AM finetuning on Hday2night/HFlickr · iHarmony4 small datasets | MSE, fMSE, fSSIM | Hurt |
| finetune SycoNet on the target dataset | SycoNet augmentation · Hday2night, HFlickr | MSE, fMSE, fSSIM | Improved |
| use only augmented data and discard original training data (ablation) | iS2AM finetuning · Hday2night, HFlickr | MSE, fMSE, fSSIM | Hurt |
Closest papers by meaning
- Progressive Painterly Image Harmonization from Low-Level Styles to High-Level Styles (AAAI 2024 · similarity 0.87)
- Training-and-Prompt-Free General Painterly Harmonization via Zero-Shot Disentenglement on Style and Content References (AAAI 2025 · similarity 0.86)
- FRIH: Fine-Grained Region-Aware Image Harmonization (AAAI 2024 · similarity 0.85)
- High-Resolution Image Harmonization with Adaptive-Interval Color Transformation (NeurIPS 2024 · similarity 0.85)
- Painterly Image Harmonization by Learning from Painterly Objects (AAAI 2024 · similarity 0.84)
- Relightful Harmonization: Lighting-aware Portrait Background Replacement (CVPR 2024 · similarity 0.84)
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.
