Research map / Detection and segmentation / Open-set and few-shot detection
Emergence of Object Segmentation in Perturbed Generative Models
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 random local shifts of the foreground relative to the background (ablation) | Layered scene generator · LSUN Car | mIoU: 0.039 -> 0.685; 0.025 -> 0.533 | Helped |
| use a shift range matched to local object shifts rather than 25% of resolution (ablation) | Layered scene generator · LSUN Car | mIoU: 0.144 -> 0.685; 0.094 -> 0.533 | Helped |
| add background contrast jitter in (0.7, 1.3) (ablation) | Layered scene generator · LSUN Car | mIoU: 0.685 -> 0.765; 0.533 -> 0.673 | Helped |
| remove random crops of real images (ablation) | Layered scene generator · LSUN Car | mIoU: 0.685 -> 0.264; 0.533 -> 0.136 | Hurt |
| increase the mask size loss coefficient to γ₁=10.0 (ablation) | Layered scene generator · LSUN Car | mIoU: 0.685 -> 0.733; 0.533 -> 0.643 | Helped |
| reduce the minimum mask size parameter to η=5% (ablation) | Layered scene generator · LSUN Car | mIoU: 0.685 -> 0.693; 0.533 -> 0.552 | No clear effect |
| use separate generators for background and foreground-with-mask rather than one shared generator (ablation) | Layered scene generator · LSUN Car | mIoU: 0.550 -> 0.685; 0.484 -> 0.533 | Helped |
| use a layered generator instead of a single-output GAN | Image generation · LSUN Car 64×64 and 128×128 | FID: 27.807 -> 31.409; 21.665 -> 30.867 | Hurt |
Closest papers by meaning
- gen2seg: Generative Models Enable Generalizable Instance Segmentation (ICLR 2026 · similarity 0.89)
- Finding an Unsupervised Image Segmenter in each of your Deep Generative Models (ICLR 2022 · similarity 0.88)
- A Generative Approach at the Instance-Level for Image Segmentation Under Limited Training Data Conditions (Student Abstract) (AAAI 2025 · similarity 0.88)
- Compositional Scene Understanding through Inverse Generative Modeling (ICML 2025 · similarity 0.87)
- Using Diffusion Priors for Video Amodal Segmentation (CVPR 2025 · similarity 0.86)
- Generative Region-Language Pretraining for Open-Ended Object Detection (CVPR 2024 · similarity 0.86)
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.