Research map / Label-efficient and robust learning / Semi-supervised learning
Transformation Consistency Regularization – A Semi-Supervised Paradigm for Image-to-Image Translation
Aamir Mustafa, Rafal K. Mantiuk
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 TCR semi-supervised training on unlabeled data (10% labeled) | Colorization network · Image colorization (Places) | PSNR: +3.1 dB | Improved |
| add TCR semi-supervised training on unlabeled data (10% labeled) | DnCNN · Image denoising | PSNR: +3.0 dB | Improved |
| add TCR semi-supervised training with 1% labeled frames | Colorization network · Big Buck Bunny movie colorization | PSNR: +6.1 dB | Improved |
| add TCR semi-supervised training with 1% labeled frames | EDSR · Elephants Dream movie super-resolution | PSNR: +5.7 dB (4x), +2.9 dB (3x) | Improved |
| add TCR loss on labeled data only (ablation) | Supervised baselines · Colorization, denoising, super-resolution | reconstruction quality (PSNR/FSIM) | Improved |
| replace geometric data augmentation with TCR (labeled only) (ablation) | Supervised baselines · Colorization, denoising, super-resolution | reconstruction quality (PSNR/FSIM) | Improved |
| add TCR so denoised images remap to the clean manifold | DnCNN (10% labeled) · Manifold validation, denoising | fraction of outputs classified as clean: 44.3% -> 97.9% | Improved |
| increase unlabeled-to-labeled ratio r per batch up to 10 (ablation) | TCR · Colorization and denoising (10% and 20% labeled) | PSNR | Improved |
| set regularization weight lambda to 10 (searched 10^-1 to 10^4) (ablation) | TCR · Colorization and denoising (10% and 20% labeled) | PSNR | Improved |
| add TCR with perceptual loss instead of L2 (ablation) | Super-resolution model · Super-resolution (perceptual loss) | PSNR | Improved |
| add TCR with L1 loss instead of L2 (ablation) | Super-resolution model · Super-resolution (L1 loss) | PSNR | Improved |
| use perceptual loss instead of pixel L2 loss (ablation) | Super-resolution model · Super-resolution | PSNR | Hurt |
Closest papers by meaning
- Semi-Supervised Learning for Few-Shot Image-to-Image Translation (CVPR 2020 · similarity 0.90)
- Adaptive Consistency Regularization for Semi-Supervised Transfer Learning (CVPR 2021 · similarity 0.88)
- Unsupervised Image-to-Image Translation with Density Changing Regularization (NeurIPS 2022 · similarity 0.87)
- Self-Supervised Dense Consistency Regularization for Image-to-Image Translation (CVPR 2022 · similarity 0.86)
- Semi-Supervised Parametric Real-World Image Harmonization (CVPR 2023 · similarity 0.86)
- Harmonic Unpaired Image-to-image Translation (ICLR 2019 · 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.
