Research map / Image generation / GANs
Generative Semantic Manipulation with Mask-Contrasting GAN
Xiaodan Liang, Hao Zhang, Liang Lin, Eric Xing
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 cycle-consistency loss to contrast-only objective (ablation) | Contrast-GAN · Cityscapes labels->photos | per-pixel accuracy: 0.53 -> 0.57 | Improved |
| add LSGAN classification loss to contrast + cycle (ablation) | Contrast-GAN · Cityscapes labels->photos | per-pixel accuracy: 0.57 -> 0.58 | Improved |
| add LSGAN classification loss to contrast + cycle (ablation) | Contrast-GAN · Cityscapes labels->photos | class IoU: 0.13 -> 0.16 | Improved |
| add LSGAN classification loss to contrast + cycle (ablation) | Contrast-GAN · Cityscapes labels->photos | per-class accuracy: 0.22 -> 0.21 | No clear effect |
| add cycle-consistency loss to contrast + classify (ablation) | Contrast-GAN · Cityscapes photos->labels | per-pixel accuracy: 0.55 -> 0.61 | Improved |
| add LSGAN classification loss to contrast + cycle (ablation) | Contrast-GAN · Cityscapes photos->labels | class IoU: 0.15 -> 0.18 | Improved |
| add the contrasting objective to cycle-consistency (CycleGAN to full model) (ablation) | CycleGAN · Cityscapes labels->photos | per-pixel accuracy: 0.52 -> 0.58 | Improved |
| add the contrasting objective to cycle-consistency (CycleGAN to full model) (ablation) | CycleGAN · Cityscapes photos->labels | per-pixel accuracy: 0.58 -> 0.61 | Improved |
| use contrast objective alone instead of CycleGAN | Contrast-GAN · Cityscapes photos->labels | per-pixel accuracy: 0.58 -> 0.54 | Hurt |
| use one shared conditional generator instead of separate generators per semantic (ablation) | Contrast-GAN · Cityscapes labels->photos | per-pixel accuracy: 0.57 -> 0.58 | No clear effect |
| use one shared conditional generator instead of separate generators per semantic (ablation) | Contrast-GAN · Cityscapes photos->labels | class IoU: 0.17 -> 0.18 | No clear effect |
| add auxiliary global discriminator D_I (ablation) | Mask Contrast-GAN · MSCOCO cat->dog (AMT) | % labeled real: 4.3% -> 4.8% | Improved |
Closest papers by meaning
- High-Resolution Image Synthesis and Semantic Manipulation With Conditional GANs (CVPR 2018 · similarity 0.93)
- Linear Semantics in Generative Adversarial Networks (CVPR 2021 · similarity 0.91)
- Navigating the GAN Parameter Space for Semantic Image Editing (CVPR 2021 · similarity 0.90)
- Editing in Style: Uncovering the Local Semantics of GANs (CVPR 2020 · similarity 0.90)
- Edge Guided GANs with Contrastive Learning for Semantic Image Synthesis (ICLR 2023 · similarity 0.89)
- Semantic Hierarchy Emerges in Deep Generative Representations for Scene Synthesis (IJCV 2021 · 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.
