Research map / Image generation / GANs
Generative Adversarial Networks for Extreme Learned Image Compression
Eirikur Agustsson, Michael Tschannen, Fabian Mentzer, Radu Timofte, Luc Van Gool
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 GAN term to MSE-only training (GC vs MSE baseline) (ablation) | GC network · Cityscapes visual quality | perceptual quality | Improved |
| use GC with C=2 or 4 channels instead of BPG | extreme-bitrate compression · Kodak, RAISE1K, Cityscapes user study | bits BPG needs for equal preference: BPG needs 95--181% more bits | Improved |
| raise bottleneck channels from C=4 to C=8 (ablation) | GC network · user studies | gain over BPG: BPG needs 21--49% more bits | Mixed |
| feed semantic label maps to discriminator during training (D+) (ablation) | GC network · Cityscapes user study | preference vs BPG: BPG needs 181% more bits (C=2) | Improved |
| synthesize repetitive classes (trees, streets, sky) in selective compression (ablation) | SC network · Cityscapes | bpp: reductions in bpp of 50% | Improved |
| synthesize complex objects (buildings, people) (ablation) | SC network · Cityscapes | visual quality | Hurt |
| add arithmetic coding of channels (ablation) | GC network · Cityscapes | bitrate: 8.8% smaller bitrates | Improved |
| use SC networks over pix2pixHD for semantics | SC network · Cityscapes validation | mean IoU | Improved |
Closest papers by meaning
- High-Fidelity Generative Image Compression (NeurIPS 2020 · similarity 0.94)
- How To Exploit the Transferability of Learned Image Compression to Conventional Codecs (CVPR 2021 · similarity 0.92)
- Self-Supervised Generative Adversarial Compression (NeurIPS 2020 · similarity 0.92)
- Deep Generative Models for Distribution-Preserving Lossy Compression (NeurIPS 2018 · similarity 0.90)
- GAN Compression: Efficient Architectures for Interactive Conditional GANs (CVPR 2020 · similarity 0.90)
- Compressing Image-to-Image Translation GANs Using Local Density Structures on Their Learned Manifold (AAAI 2024 · similarity 0.88)
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.
