Research map / Image generation / GANs
Wavelet Knowledge Distillation: Towards Efficient Image-to-Image Translation
Check 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 |
|---|---|---|---|
| Distill only high-frequency wavelet bands instead of matching whole generated images (ablation) | CycleGAN student · Horse→Zebra and Zebra→Horse | FID: 85.04 -> 77.04; 152.67 -> 146.01 | Helped |
| Distill only the low-frequency band (ablation) | CycleGAN student · Horse→Zebra and Zebra→Horse | FID: 85.04 -> 96.11; 152.67 -> 163.06 | Hurt |
| Distill both low- and high-frequency bands rather than high-frequency bands alone (ablation) | CycleGAN student · Horse→Zebra and Zebra→Horse | FID: 77.04 -> 81.81; 146.01 -> 148.09 | Hurt |
| Add wavelet KD to the compressed generator | Pix2Pix and Pix2PixHD students · Edges→Shoes | FID: 85.06 -> 80.13; 44.64 -> 42.53 | Helped |
| Add wavelet KD to the compressed generator | CycleGAN student · Horse→Zebra and Zebra→Horse | FID: 70.54 -> 61.65; 141.86 -> 138.84 | Helped |
| Add wavelet KD to the compressed Pix2Pix generator | Pix2Pix student · Cityscapes | mIoU: 41.35 -> 42.93 | Helped |
| Add Li and Lin feature distillation alongside wavelet KD | CycleGAN student · Horse→Zebra and Zebra→Horse | FID: 77.04 -> 76.40; 146.01 -> 145.96; 61.65 -> 60.13; 138.84 -> 138.52 | Helped |
| Use Hinton image-level KD instead of no KD | Pix2Pix student · Edges→Shoes | FID: 85.06 -> 86.97 | Hurt |
| Use Hinton image-level KD instead of no KD | Pix2PixHD student · Edges→Shoes | FID: 44.64 -> 45.31 | Hurt |
| Compress the discriminator to 4.01× while retaining a 15.81× compressed generator (ablation) | CycleGAN with wavelet KD · Horse→Zebra and Zebra→Horse | FID | Helped |
| Use a teacher-assistant KD scheme instead of teacher-student KD (ablation) | CycleGAN student with wavelet KD · Horse→Zebra and Zebra→Horse | FID: 77.04 -> 78.53; 146.01 -> 148.03 | No clear effect |
| Use self-distillation instead of teacher-student KD (ablation) | CycleGAN student with wavelet KD · Horse→Zebra and Zebra→Horse | FID: 77.04 -> 83.51; 146.01 -> 151.74 | Hurt |
Closest papers by meaning
- Distilling Portable Generative Adversarial Networks for Image Translation (AAAI 2020 · similarity 0.91)
- KD-DLGAN: Data Limited Image Generation via Knowledge Distillation (CVPR 2023 · similarity 0.90)
- E$^2$GAN: Efficient Training of Efficient GANs for Image-to-Image Translation (ICML 2024 · similarity 0.87)
- Small Scale Data-Free Knowledge Distillation (CVPR 2024 · similarity 0.85)
- Unlocking Pre-trained Image Backbones for Semantic Image Synthesis (CVPR 2024 · similarity 0.85)
- Compressing Image-to-Image Translation GANs Using Local Density Structures on Their Learned Manifold (AAAI 2024 · similarity 0.85)
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.