Research map / Image generation / Autoregressive and latent generation
Edge Guided GANs with Contrastive Learning for Semantic Image Synthesis
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 an edge generator to the image-generation branch (ablation) | ECGAN · Cityscapes | mIoU: 58.6 -> 60.2 | Helped |
| Add attention-guided edge transfer (ablation) | ECGAN with edge generation · Cityscapes | mIoU: 60.2 -> 61.5 | Helped |
| Add the semantic preserving module (ablation) | ECGAN with edge transfer · Cityscapes | mIoU: 61.5 -> 64.5 | Helped |
| Add the label generator and similarity loss (ablation) | ECGAN with semantic preservation · Cityscapes | mIoU: 64.5 -> 66.8 | Helped |
| Add pixel-wise contrastive learning and class-specific pixel generation (ablation) | ECGAN with similarity loss · Cityscapes | mIoU: 66.8 -> 72.2 | Helped |
| Replace Canny edge extraction with HED (ablation) | ECGAN · Cityscapes | FID: 56.7 (HED); Canny result 44.5 | Hurt |
| Use one multi-modality discriminator instead of separate edge and image discriminators | ECGAN · image and edge synthesis | performance | No clear effect |
| Increase the number of layouts used for contrastive loss beyond 8 | ECGAN · training across semantic layouts | performance | No clear effect |
| Use ECGAN instead of OASIS | semantic image synthesis · Cityscapes | FID: 47.7 -> 44.5 | Helped |
| Use ECGAN instead of OASIS | semantic image synthesis · ADE20K | FID: 28.3 -> 25.8 | Helped |
| Use ECGAN instead of OASIS | semantic image synthesis · COCO-Stuff | FID: 17.0 -> 15.7 | Helped |
| Use ECGAN instead of OASIS | semantic image synthesis · Cityscapes small objects | overall mIoU: 26.0 -> 28.8 | Helped |
Closest papers by meaning
- Learning to Predict Layout-to-image Conditional Convolutions for Semantic Image Synthesis (NeurIPS 2019 · similarity 0.91)
- You Only Need Adversarial Supervision for Semantic Image Synthesis (ICLR 2021 · similarity 0.91)
- Linear Semantics in Generative Adversarial Networks (CVPR 2021 · similarity 0.91)
- Unlocking Pre-trained Image Backbones for Semantic Image Synthesis (CVPR 2024 · similarity 0.90)
- High-Resolution Image Synthesis and Semantic Manipulation With Conditional GANs (CVPR 2018 · similarity 0.89)
- Semantic Pyramid for Image Generation (CVPR 2020 · 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.