Research map / Detection and segmentation / CNN features
GMMSeg: Gaussian Mixture based Generative Semantic Segmentation 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 |
|---|---|---|---|
| train GMM online with discriminative features instead of post-hoc GMM on softmax features (ablation) | GMMSeg-DeepLabV3+ · ADE20K val | mIoU: 31.6 -> 46.0 | Helped |
| learn GMM generatively via EM rather than end-to-end discriminatively (ablation) | GMMSeg-DeepLabV3+ · Fishyscapes Lost&Found val | AP: 17.68 -> 43.47 | Helped |
| learn GMM generatively via EM rather than end-to-end discriminatively (ablation) | GMMSeg-DeepLabV3+ · Cityscapes val | mIoU: 81.0 -> 81.1 | No clear effect |
| replace vanilla EM with Sinkhorn EM (1 loop) (ablation) | GMMSeg-DeepLabV3+ · ADE20K val | mIoU: 42.7 -> 46.0 | Helped |
| more EM loops per iteration (1 to 10) with Sinkhorn EM (ablation) | GMMSeg-DeepLabV3+ · ADE20K val | mIoU: 46.0 -> 46.0 | No clear effect |
| increase Gaussians per class from 1 to 5 (ablation) | GMMSeg-DeepLabV3+ · ADE20K val | mIoU: 44.2 -> 46.0 | Helped |
| increase Gaussians per class from 5 to 15 (ablation) | GMMSeg-DeepLabV3+ · ADE20K val | mIoU: 46.0 -> 45.7 | No clear effect |
| add external memory of 32K samples (ablation) | GMMSeg-DeepLabV3+ · ADE20K val | mIoU: 40.3 -> 46.0 | Helped |
| replace softmax classifier with GMM generative classifier | DeepLabV3+ ResNet101 · ADE20K val | mIoU: 45.5 -> 46.7 | Helped |
| replace softmax classifier with GMM generative classifier | SegFormer MiT-B5 · Cityscapes val | mIoU: 82.0 -> 82.6 | Helped |
| replace softmax classifier with GMM classifier | Mask2Former Swin-L · COCO-Stuff test | mIoU: 51.0 -> 52.0 | Helped |
Closest papers by meaning
- CGMGM: A Cross-Gaussian Mixture Generative Model for Few-Shot Semantic Segmentation (AAAI 2024 · similarity 0.89)
- RankSEG-RMA: An Efficient Segmentation Algorithm via Reciprocal Moment Approximation (NeurIPS 2025 · similarity 0.86)
- On the Relaxation of Conditional Independence Assumption for Image Segmentation (NeurIPS 2026 · similarity 0.86)
- ARGenSeg: Image Segmentation with Autoregressive Image Generation Model (NeurIPS 2025 · similarity 0.85)
- Unveiling the Knowledge of CLIP for Training-Free Open-Vocabulary Semantic Segmentation (AAAI 2025 · similarity 0.85)
- Hierarchy-Agnostic Unsupervised Segmentation: Parsing Semantic Image Structure (NeurIPS 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.