Research map / Image generation / GANs
GanOrCon: Are Generative Models Useful for Few-Shot Segmentation?
Check 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 |
|---|---|---|---|
| Use contrastive CONV representations with distillation rather than the strongest listed prior method | Few-shot part segmentation · Face34, Face8, Car20 | mIoU: 0.5365 ± 0.0043 -> 0.5406 ± 0.0015; 0.7001 -> 0.7048 ± 0.0028; 0.6840 ± 0.0322 -> 0.6956 ± 0.0178 | Helped |
| Use contrastive CONV representations with distillation rather than DatasetGAN | Few-shot part segmentation · Cat16 | mIoU: 0.3126 ± 0.0071 -> 0.3101 ± 0.0042 | Hurt |
| Add distillation to the StyleGAN + MLP segmentor (ablation) | StyleGAN + MLP · Face34 | mIoU: 0.4301 -> 0.5365 | Helped |
| Add distillation to MoCo + MLP (ablation) | MoCo + MLP · Face34 | mIoU: 0.5162 -> 0.5341 | Helped |
| Add distillation to MoCo + CONV (ablation) | MoCo + CONV · Face34 | mIoU: 0.5414 -> 0.5406 | Hurt |
| Use a CONV projector instead of an MLP projector (ablation) | MoCo few-shot segmentor · Face34 | mIoU: 0.5162 -> 0.5414 | Helped |
| Use MoCo rather than SimSiam (ablation) | ResNet50 contrastive backbone with MLP projector · Face34 | mIoU: 0.5038 -> 0.5162 | Helped |
| Use MoCo rather than SimSiam (ablation) | ResNet50 contrastive backbone with CONV projector · Face34 | mIoU: 0.5297 -> 0.5414 | Helped |
| Use MoCo pretraining rather than ImageNet classification pretraining (ablation) | ResNet50 with MLP projector · Face34 | mIoU: 0.1893 -> 0.5162 | Helped |
| Pretrain MoCo on CelebA rather than FFHQ (ablation) | ResNet50 with MLP projector · Face34 | mIoU: 0.5023 -> 0.5162 | Helped |
| Train MoCo on StyleGAN-generated images instead of real images (ablation) | MoCo ResNet50 · Face34 | mIoU: 0.5162 -> 0.3868 | Hurt |
| Train MoCo on StyleGAN-generated images instead of real images (ablation) | MoCo ResNet50 · Car20 | mIoU: 0.6143 -> 0.4266 | Hurt |
Closest papers by meaning
- Repurposing GANs for One-Shot Semantic Part Segmentation (CVPR 2021 · similarity 0.90)
- Training Generative Adversarial Networks from Incomplete Observations using Factorised Discriminators (ICLR 2020 · similarity 0.87)
- LARGE: Latent-Based Regression Through GAN Semantics (CVPR 2022 · similarity 0.87)
- PandA: Unsupervised Learning of Parts and Appearances in the Feature Maps of GANs (ICLR 2023 · similarity 0.87)
- A Generative Approach at the Instance-Level for Image Segmentation Under Limited Training Data Conditions (Student Abstract) (AAAI 2025 · similarity 0.87)
- Linear Semantics in Generative Adversarial Networks (CVPR 2021 · similarity 0.87)
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.