Research map / Detection and segmentation / Open-set and few-shot detection
Generative Semantic Segmentation
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 |
|---|---|---|---|
| Replace random beta initialization with maximal-distance initialization (ablation) | GSS-FF maskige reconstruction · ADE20K val | mIoU: 62.83 -> 84.31 | Helped |
| Use a nonlinear inverse transformation instead of a linear inverse (ablation) | GSS maskige reconstruction · ADE20K val | mIoU: 84.31 -> 86.10 | Helped |
| Replace the CNN inverse transformation with a shifted-window Transformer (ablation) | GSS-FT maskige reconstruction · ADE20K val | mIoU: 86.10 -> 87.73 | Helped |
| Train beta automatically while retaining a linear inverse transformation (ablation) | GSS-FF maskige reconstruction · ADE20K val | mIoU: 84.31 -> 84.37 | No clear effect |
| Jointly optimize the maskige transform and its inverse (ablation) | GSS maskige reconstruction · ADE20K val | mIoU: 84.31 -> 36.11 | Hurt |
| Add unlabeled-area auxiliary pseudo-labeling (ablation) | GSS latent prior learning · ADE20K val | mIoU: 40.64 -> 43.72 | Helped |
| Increase latent mask resolution from 1/8 to 1/4 (ablation) | GSS latent prior learning with unlabeled-area auxiliary · ADE20K val | mIoU: 43.72 -> 43.98 | Helped |
| Add Multi-Level Aggregation (ablation) | GSS latent prior learning · ADE20K val | mIoU: 43.98 -> 46.29 | Helped |
| Represent segmentation masks as RGB maskige for VQGAN reconstruction (ablation) | VQGAN reconstruction · Cityscapes val | mIoU: 82.16 -> 75.09 | Hurt |
| Use DALL·E-style pretrained VQVAE rather than VQGAN with maskige (ablation) | Maskige reconstruction · ADE20K val | mIoU: 42.70 -> 87.73 | Helped |
| Add misaligned channel starting points to arithmetic colorization (ablation) | GSS-FF maskige encoding · ADE20K val | mIoU: 85.99 -> 86.12 | Helped |
| Add random channel-specific factors to colorization (ablation) | GSS-FF maskige encoding · ADE20K val | mIoU: 86.12 -> 87.42 | Helped |
Closest papers by meaning
- GenMask: Adapting DiT for Segmentation via Direct Mask Generation (CVPR 2026 · similarity 0.89)
- MatchMask: Mask-Centric Generative Data Augmentation for Label-Scarce Semantic Segmentation (CVPR 2026 · similarity 0.89)
- A Generative Approach at the Instance-Level for Image Segmentation Under Limited Training Data Conditions (Student Abstract) (AAAI 2025 · similarity 0.88)
- Generative Medical Segmentation (AAAI 2025 · similarity 0.88)
- MaskViM: Domain Generalized Semantic Segmentation with State Space Models (AAAI 2025 · similarity 0.87)
- Unleashing the Potential of the Diffusion Model in Few-shot Semantic Segmentation (NeurIPS 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.