Research map / Image generation / Editing and style transfer
SinIR: Efficient General Image Manipulation with Single Image Reconstruction
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| reconstruction objective instead of GAN objective (plus multi-scale and pixel shuffling) | SinIR vs SinGAN · Training time, 500px images | training time speedup: 33.49x faster (8m 0s vs 267m 57s) | Helped |
| random pixel shuffling omitted from training (ablation) | SinIR · Manipulation quality (qualitative) | visual quality | Hurt |
| random pixel shuffling at 0.05% to 5% of pixels (ablation) | SinIR · Photo-realistic style transfer / manipulation | visual quality | Helped |
| increase shuffled pixel percentage during training (ablation) | SinIR super-resolution · 4X super-resolution BSD100 | perception-distortion tradeoff (NIQE vs RMSE): 5e-2 RMSE 16.56 NIQE 3.647 vs 5e-4 RMSE 17.23 NIQE 3.402 | Mixed |
| SinIR vs SinGAN for 4X super-resolution perceptual quality | SinIR (5e-4) · 4X super-resolution BSD100 | NIQE: 3.402 vs 3.709 for SinGAN | Helped |
| SinIR vs SinGAN in user study | SinIR · Photo-realistic style transfer user study | preference rate: 90.00% vs 6.75% | Helped |
| SinIR vs SinGAN in user study | SinIR · Artistic style transfer user study | preference rate: 76.95% vs 18.70% | Helped |
| SinIR vs SinGAN in user study | SinIR · Editing user study | preference rate: 51.25% vs 37.00% | Helped |
| SinIR vs SinGAN in user study | SinIR · Harmonization user study | preference rate: 56.90% vs 33.33% | Helped |
Closest papers by meaning
- Learning to Manipulate Artistic Images (AAAI 2024 · similarity 0.87)
- ReGANIE: Rectifying GAN Inversion Errors for Accurate Real Image Editing (AAAI 2023 · similarity 0.87)
- Drop the GAN: In Defense of Patches Nearest Neighbors As Single Image Generative Models (CVPR 2022 · similarity 0.87)
- PetsGAN: Rethinking Priors for Single Image Generation (AAAI 2022 · similarity 0.86)
- E$^2$GAN: Efficient Training of Efficient GANs for Image-to-Image Translation (ICML 2024 · similarity 0.84)
- Gradual Residuals Alignment: A Dual-Stream Framework for GAN Inversion and Image Attribute Editing (AAAI 2024 · similarity 0.83)
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.