Research map / Image generation / GANs
GIQA: Generated Image Quality Assessment
Shuyang Gu, Jianmin Bao, Dong Chen, Fang Wen
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 multiple binary classifiers instead of direct regression of pseudo label (ablation) | MBC-GIQA · LGIQA-FFHQ | pairwise accuracy: 0.678 -> 0.731 | Improved |
| use multiple binary classifiers instead of a single real/fake classifier (ablation) | MBC-GIQA · LGIQA-FFHQ | pairwise accuracy: 0.663 -> 0.731 | Improved |
| use Gaussian mixture instead of a single Gaussian (ablation) | GMM-GIQA · LGIQA-FFHQ | pairwise accuracy: 0.620 -> 0.764 | Improved |
| use GMM-GIQA rather than SGM-GIQA on LSUN-cat (ablation) | GMM-GIQA · LGIQA-LSUN-cat | pairwise accuracy: 0.829 -> 0.846 | Improved |
| add OHEM weighting on low-quality samples in discriminator (ablation) | StyleGAN · FFHQ (256x256) | FID: 17.35 -> 16.89 | Improved |
| add OHEM weighting on low-quality samples in discriminator (ablation) | StyleGAN · LSUN-cat (256x256) | FID: 18.67 -> 18.12 | Improved |
| add image picker to discard low-quality samples (ablation) | StyleGAN+OHEM · LSUN-cat (256x256) | FID: 18.12 -> 16.25 | Improved |
| add image picker to discard low-quality samples (ablation) | StyleGAN+OHEM · FFHQ (256x256) | FID: 16.89 -> 16.68 | Improved |
| use image picker at remaining rate 0.9 versus truncation rate 0.9 | StyleGAN · LSUN-cat | FID: 18.05 -> 16.65 | Improved |
| increase number of binary classifiers beyond 8 (ablation) | MBC-GIQA · LGIQA-FFHQ | pairwise accuracy: 0.731 -> 0.717 | Hurt |
| use training patch size 192 instead of 256 (ablation) | MBC-GIQA · LGIQA-FFHQ | pairwise accuracy: 0.690 -> 0.731 | Improved |
| increase Gaussian components from 70 to 100 (ablation) | GMM-GIQA · LGIQA-FFHQ | pairwise accuracy: 0.764 -> 0.753 | Hurt |
Closest papers by meaning
- Quality Aware Generative Adversarial Networks (NeurIPS 2019 · similarity 0.89)
- How good is my GAN? (ECCV 2018 · similarity 0.88)
- Progressive Growing of GANs for Improved Quality, Stability, and Variation (ICLR 2018 · similarity 0.88)
- Are GANs Created Equal? A Large-Scale Study (NeurIPS 2018 · similarity 0.88)
- Revisiting the Evaluation of Image Synthesis with GANs (NeurIPS 2023 · similarity 0.87)
- Anomaly Score: Evaluating Generative Models and Individual Generated Images based on Complexity and Vulnerability (CVPR 2024 · similarity 0.86)
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.
