Research map / Image generation / GANs
Teachers Do More Than Teach: Compressing Image-to-Image Models
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 |
|---|---|---|---|
| Compress CycleGAN using CAT's teacher-based one-step pruning and distillation | CycleGAN · Horse→Zebra | FID: 61.53 -> 53.48 | Helped |
| Compress Pix2pix using CAT's teacher-based one-step pruning and distillation | Pix2pix · Cityscapes | mIoU: 42.06 -> 42.65 | Helped |
| Compress Pix2pix using CAT's teacher-based one-step pruning and distillation | Pix2pix · Map→Aerial photo | FID: 47.76 -> 45.63 | Helped |
| Compress GauGAN to CAT-A's computation budget | GauGAN · Cityscapes | mIoU: 62.18 -> 62.35 | Helped |
| Compress GauGAN more aggressively to CAT-B | GauGAN · Cityscapes | mIoU: 62.18 -> 54.71 | Hurt |
| Use CAT instead of the original generator | CycleGAN · Horse→Zebra | KID: 0.020±0.002 -> 0.015±0.001 | Helped |
| Use CAT instead of the original generator | Pix2pix · Map→Aerial | KID: 0.154±0.010 -> 0.012±0.002 | Helped |
| Use CAT-A instead of the original generator | GauGAN · Cityscapes | KID: 0.026±0.003 -> 0.014±0.002 | Helped |
| Remove knowledge distillation (ablation) | Pix2pix student · Cityscapes | mIoU: 39.39 -> 42.65 | Hurt |
| Use MSE feature distillation with loss weight 0.5 instead of KDKA (ablation) | Pix2pix student · Cityscapes | mIoU: 39.83 -> 42.65 | Hurt |
| Use MSE feature distillation with loss weight 1.0 instead of KDKA (ablation) | Pix2pix student · Cityscapes | mIoU: 39.76 -> 42.65 | Hurt |
| Use kernel-alignment similarity for feature distillation rather than MSE (ablation) | Pix2pix student · Cityscapes | mIoU: 39.83 -> 42.65 | Helped |
Closest papers by meaning
- Distilling Portable Generative Adversarial Networks for Image Translation (AAAI 2020 · similarity 0.90)
- GAN Compression: Efficient Architectures for Interactive Conditional GANs (CVPR 2020 · similarity 0.90)
- Discriminator-Cooperated Feature Map Distillation for GAN Compression (CVPR 2023 · similarity 0.90)
- Content-Aware GAN Compression (CVPR 2021 · similarity 0.89)
- Compressing Image-to-Image Translation GANs Using Local Density Structures on Their Learned Manifold (AAAI 2024 · similarity 0.88)
- Hybrid Data-Free Knowledge Distillation (AAAI 2025 · similarity 0.87)
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.