Research map / Image generation / GANs
Local and Global GANs With Semantic-Aware Upsampling for Image Generation
Hao Tang, Ling Shao, Philip H.S. Torr, Nicu Sebe
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 local generator (addition) (ablation) | LGGAN global only · Cityscapes | mIoU / FID: 62.3 -> 64.6; 71.8 -> 66.1 | Improved |
| generate local result by convolution instead of addition (ablation) | LGGAN · Cityscapes | mIoU / FID: 64.6 -> 65.8; 66.1 -> 65.6 | Improved |
| add class discriminative loss (ablation) | LGGAN B3 · Cityscapes | mIoU / FID: 65.8 -> 67.0; 65.6 -> 61.3 | Improved |
| add weight map fusion (ablation) | LGGAN B4 · Cityscapes | mIoU / FID: 67.0 -> 68.4; 61.3 -> 57.7 | Improved |
| replace upsampling with SAU | GauGAN · Cityscapes | FID: 71.8 -> 48.3 | Improved |
| replace upsampling with SAU | SEAN · Cityscapes | FID: 61.8 -> 50.3 | Improved |
| replace upsampling with SAU | MaskGAN · Cityscapes | FID: 65.8 -> 51.3 | Improved |
| replace upsampling with SAU | LGGAN · Cityscapes | mIoU: 68.4 -> 67.7 | No clear effect |
| replace upsampling with SAU | GauGAN · COCO-Stuff | mIoU / FID: 37.4 -> 39.0; 22.6 -> 20.1 | Improved |
| use SAU instead of bilinear upsampling (ablation) | GauGAN · Cityscapes | FID: 52.9 -> 48.3 | Improved |
| use SAU instead of spatial attention or involution (ablation) | GauGAN · Cityscapes | FID: 56.2 -> 48.3; 58.8 -> 48.3 | Improved |
| use pixel shuffle instead of bilinear upsampling (ablation) | GauGAN · Cityscapes | FID: 52.9 -> 59.1 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- SEAN — SEAN: Image Synthesis With Semantic Region-Adaptive Normalization (CVPR 2020) · better on 2 of 2 shared results
- MaskGAN — MaskGAN: Better Text Generation via Filling in the _______ (ICLR 2018) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Local Class-Specific and Global Image-Level Generative Adversarial Networks for Semantic-Guided Scene Generation (CVPR 2020 · similarity 0.93)
- Collaging Class-Specific GANs for Semantic Image Synthesis (ICCV 2021 · similarity 0.92)
- Semantic Pyramid for Image Generation (CVPR 2020 · similarity 0.92)
- Linear Semantics in Generative Adversarial Networks (CVPR 2021 · similarity 0.91)
- Unlocking Pre-trained Image Backbones for Semantic Image Synthesis (CVPR 2024 · similarity 0.91)
- Edge Guided GANs with Contrastive Learning for Semantic Image Synthesis (ICLR 2023 · 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.
