Research map / Image generation / GANs
Collaging Class-Specific GANs for Semantic Image Synthesis
Yuheng Li, Yijun Li, Jingwan Lu, Eli Shechtman, Yong Jae Lee, Krishna Kumar Singh
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 additional training data for class-specific generator (ablation) | CollageGAN · car (Cityscapes) | FID: 36.71 -> 30.42 | Improved |
| add additional training data for class-specific generator (ablation) | CollageGAN · person (Cityscapes) | FID: 88.47 -> 82.34 | Improved |
| add class-specific generators on top of base model (no extra data) (ablation) | Base model · car (Cityscapes) | FID: 44.50 -> 36.71 | Improved |
| add class-specific generators on top of base model | Base model · pillow (Bedroom) | FID: 127.67 -> 136.79 | Hurt |
| add class-specific generators on top of base model | Base model · chest, chair, lamp, table (Bedroom) | FID: 146.15 -> 132.12 | Improved |
| add top-down pathway for multi-resolution encoder features (ablation) | Base model · ADE20K Bedroom | FID: 40.88 -> 34.41 | Improved |
| add context input to class-specific generator (ablation) | CollageGAN · lamp/face qualitative | visual consistency | Improved |
| replace OASIS-style base with class-specific compositing for alignment | Base model · Cityscapes user alignment vs OASIS | user preference %: 18.2 -> 46.8 | Improved |
| use class-specific generators for foreground objects | Base model · Cityscapes foreground object classification | top-1 accuracy %: 68.80 -> 82.64 | Improved |
| use CollageGAN versus OASIS in realism study | OASIS · Cityscapes user study | user preference %: 35.2% -> 83.6% | Improved |
Closest papers by meaning
- Local and Global GANs With Semantic-Aware Upsampling for Image Generation (TPAMI 2022 · similarity 0.92)
- Linear Semantics in Generative Adversarial Networks (CVPR 2021 · similarity 0.92)
- Local Class-Specific and Global Image-Level Generative Adversarial Networks for Semantic-Guided Scene Generation (CVPR 2020 · similarity 0.92)
- You Only Need Adversarial Supervision for Semantic Image Synthesis (ICLR 2021 · similarity 0.91)
- Unlocking Pre-trained Image Backbones for Semantic Image Synthesis (CVPR 2024 · similarity 0.91)
- Conditional Image Synthesis with Auxiliary Classifier GANs (ICML 2017 · similarity 0.91)
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.
