Research map / Image generation / GANs
Co-Evolutionary Compression for Unpaired Image Translation
Han Shu, Yunhe Wang, Xu Jia, Kai Han, Hanting Chen, Chunjing Xu, Qi Tian, Chang Xu
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 |
|---|---|---|---|
| use discriminator-aware loss instead of ThiNet reconstruction-based pruning | compressed CycleGAN generator · Cityscapes FCN-score | mean class IoU: 0.054 -> 0.131 | Improved |
| use discriminator-aware loss instead of ThiNet reconstruction-based pruning | compressed CycleGAN generator · Cityscapes FCN-score | mean pixel accuracy: 0.218 -> 0.542 | Improved |
| use co-evolutionary compression instead of ThiNet | compressed CycleGAN generator · horse2zebra | FID: 189.28 -> 96.15 | Improved |
| use co-evolutionary compression instead of ThiNet | compressed CycleGAN generator · zebra2horse | FID: 184.88 -> 157.90 | Improved |
| use co-evolutionary compression instead of ThiNet | compressed CycleGAN generator · summer2winter | FID: 81.06 -> 78.58 | No clear effect |
| use co-evolutionary compression instead of ThiNet | compressed CycleGAN generator · winter2summer | FID: 80.17 -> 79.16 | No clear effect |
| compress with co-evolutionary method instead of original uncompressed generator | CycleGAN generator · horse2zebra | FID: 74.04 -> 96.15 | Hurt |
| compress with co-evolutionary method instead of original uncompressed generator | CycleGAN generator · Cityscapes FCN-score | mean class IoU: 0.121 -> 0.131 | No clear effect |
| compress both generators jointly with co-evolution instead of separately (ablation) | evolutionary compression with L_DisA · horse2zebra / summer2winter | visual quality | Improved |
| use discriminator-aware loss instead of generator-aware loss (ablation) | co-evolutionary compression · horse2zebra / summer2winter | visual quality | Improved |
| increase gamma (ablation) | co-evolutionary compression · horse2zebra | visual quality vs compression | Mixed |
| remove filters physically via compression | compressed CycleGAN generator · horse2zebra CPU runtime | runtime per image (ms): 2,260 -> 730 | Improved |
Closest papers by meaning
- Distilling Portable Generative Adversarial Networks for Image Translation (AAAI 2020 · similarity 0.91)
- Compressing Image-to-Image Translation GANs Using Local Density Structures on Their Learned Manifold (AAAI 2024 · similarity 0.90)
- UGC: Unified GAN Compression for Efficient Image-to-Image Translation (ICCV 2023 · similarity 0.90)
- Revisiting Discriminator in GAN Compression: A Generator-discriminator Cooperative Compression Scheme (NeurIPS 2021 · similarity 0.89)
- GAN Compression: Efficient Architectures for Interactive Conditional GANs (CVPR 2020 · similarity 0.89)
- Generative Adversarial Networks for Extreme Learned Image Compression (ICCV 2019 · similarity 0.89)
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.
