Research map / Image generation / Autoregressive and latent generation
LEMaRT: Label-Efficient Masked Region Transform for Image Harmonization
Sheng Liu, Cong Phuoc Huynh, Cong Chen, Maxim Arap, Raffay Hamid
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 |
|---|---|---|---|
| pre-train SwinIH with LEMaRT before iHarmony4 training (ablation) | SwinIH · iHarmony4 all | PSNR: 37.0 -> 39.0 | Improved |
| pre-train ViT with LEMaRT (ablation) | ViT · iHarmony4 all | PSNR: 35.7 -> 38.4 | Improved |
| pre-train ResNet with LEMaRT (ablation) | ResNet · iHarmony4 all | PSNR: 34.6 -> 36.3 | Improved |
| pre-train HRNet with LEMaRT (ablation) | HRNet · iHarmony4 all | PSNR: 33.2 -> 35.3 | Improved |
| pre-train HT+ with LEMaRT (ablation) | HT+ · iHarmony4 all | PSNR: 37.7 -> 38.9 | Improved |
| reduce then enlarge feature-map resolution (ablation) | SwinIH · iHarmony4 all | PSNR: 37.0 -> 36.3 | Hurt |
| replace the final global attention layer with Swin attention (ablation) | SwinIH · iHarmony4 all | PSNR: 37.0 -> 37.0 | No clear effect |
| replace the mixed attention architecture with Swin-T (ablation) | SwinIH · iHarmony4 all | PSNR: 37.0 -> 36.1 | Hurt |
| use fixed grid rather than random mask generation (ablation) | LEMaRT pretraining · iHarmony4 all | PSNR: 39.0 -> 37.1 | Hurt |
| use block rather than random mask generation (ablation) | LEMaRT pretraining · iHarmony4 all | PSNR: 39.0 vs 38.6 | No clear effect |
| vary foreground mask ratio from 30% to 70% (ablation) | LEMaRT pretraining · iHarmony4 all | PSNR: 38.8; 39.0; 39.0 | No clear effect |
| use LEMaRT-pretrained SwinIH rather than SCS-Co | iHarmony4 · iHarmony4 all | PSNR: 38.8 -> 39.8 | Improved |
Closest papers by meaning
- Semantic-Guided Multi-Mask Image Harmonization (ECCV 2022 · similarity 0.87)
- FRIH: Fine-Grained Region-Aware Image Harmonization (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)
- Prior-guided Hierarchical Harmonization Network for Efficient Image Dehazing (AAAI 2025 · similarity 0.84)
- Progressive Painterly Image Harmonization from Low-Level Styles to High-Level Styles (AAAI 2024 · similarity 0.84)
- Frequency-Guided Masking for Enhanced Vision Self-Supervised Learning (ICLR 2025 · 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.
