Research map / Label-efficient and robust learning / Semi-supervised learning
Learning to Annotate Part Segmentation with Gradient Matching
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 |
|---|---|---|---|
| learn annotator labels by matching gradients on generated and manually labeled data | Gradient-matching annotator learning · Cat-16 | mIoU: 32.63 ± 0.68 -> 33.89 ± 0.43 | Helped |
| learn annotator labels by matching gradients on generated and manually labeled data | Gradient-matching annotator learning · Face-34 | mIoU: 54.55 ± 0.25 -> 52.58 ± 0.61 | No clear effect |
| learn annotator labels by matching gradients on generated and manually labeled data | Gradient-matching annotator learning · Car-20 | mIoU: 67.53 ± 2.58 -> 63.55 ± 2.25 | Hurt |
| use StyleGAN2 rather than StyleGAN as the generator | Gradient-matching annotator learning · Face-34 | mIoU: 52.58 ± 0.61 -> 55.10 ± 0.39 | Helped |
| use StyleGAN2 rather than StyleGAN as the generator | Gradient-matching annotator learning · Cat-16 | mIoU: 33.89 ± 0.43 -> 33.56 ± 0.17 | No clear effect |
| use StyleGAN2 rather than StyleGAN as the generator | Gradient-matching annotator learning · Car-20 | mIoU: 63.55 ± 2.25 -> 61.21 ± 2.07 | Hurt |
| replace gradient matching with one-step MAML | Annotator learning · Face-34 | mIoU: 55.10 ± 0.39 -> 33.82 ± 0.24 | Hurt |
| replace gradient matching with pseudo-labeling from a trained segmentation model | Annotator learning · Car-20 | mIoU: 61.21 ± 2.07 -> 12.37 ± 0.91 | Hurt |
| replace gradient matching with GAN inversion features and supervised annotator training | Annotator learning · Cat-16 | mIoU: 33.56 ± 0.17 -> 15.14 ± 0.27 | Hurt |
| update the segmentation network more frequently during annotator learning (ablation) | Gradient-matching annotator · Face-34 and Cat-16 | annotator mIoU | Helped |
| skip bottom blocks and match gradients only on Res2 through Head instead of all blocks (ablation) | Gradient-matching annotator · Cat-16 | mIoU: 33.71 ± 0.77 -> 34.38 ± 0.49 | No clear effect |
| match gradients only on the segmentation head (ablation) | Gradient-matching annotator · Cat-16 | mIoU: 33.71 ± 0.77 -> 33.56 ± 0.17 | No clear effect |
Closest papers by meaning
- Repurposing GANs for One-Shot Semantic Part Segmentation (CVPR 2021 · similarity 0.90)
- DatasetGAN: Efficient Labeled Data Factory With Minimal Human Effort (CVPR 2021 · similarity 0.88)
- Semantic Segmentation With Generative Models: Semi-Supervised Learning and Strong Out-of-Domain Generalization (CVPR 2021 · similarity 0.88)
- GANSeg: Learning To Segment by Unsupervised Hierarchical Image Generation (CVPR 2022 · similarity 0.88)
- SCOPS: Self-Supervised Co-Part Segmentation (CVPR 2019 · similarity 0.87)
- Learning Hierarchical Image Segmentation For Recognition and By Recognition (ICLR 2024 · similarity 0.85)
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.