Research map / Detection and segmentation / CNN features
Per-Pixel Classification is Not All You Need for 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 |
|---|---|---|---|
| switch from per-pixel to mask classification (fixed matching) (ablation) | PerPixelBaseline+ · ADE20K val | mIoU: 41.9 -> 43.7 | Helped |
| use bipartite instead of fixed matching (ablation) | MaskFormer · ADE20K val | PQSt: 30.3 -> 33.4 | Helped |
| mask classification vs per-pixel | R101 model · Cityscapes (19 classes) | mIoU: 78.5 -> 78.5 | No clear effect |
| mask classification vs per-pixel | R50 model · ADE20K-Full (847 classes) | mIoU: 13.9 -> 17.4 | Helped |
| MaskFormer vs Swin-UperNet | Swin-L · ADE20K val | mIoU (m.s.): 53.5 -> 55.6 | Helped |
| mask-based instead of box-based matching (ablation) | MaskFormer · COCO panoptic val | PQ: 43.7 -> 46.5 | Helped |
| replace DETR per-query mask head with per-image FPN head (box matching) (ablation) | DETR · COCO panoptic val | PQ: 43.4 -> 43.7 | No clear effect |
| increase queries from 100 to 1000 (ablation) | MaskFormer · ADE20K val | mIoU: 44.5 -> 35.4 | Hurt |
| reduce queries from 100 to 20 (ablation) | MaskFormer · ADE20K val | mIoU: 44.5 -> 42.9 | Hurt |
| train PerPixelBaseline+ with focal+dice losses (ablation) | PerPixelBaseline+ · ADE20K val | mIoU | No clear effect |
| MaskFormer inference vs DETR post-processing (ablation) | MaskFormer R50 · COCO panoptic val | PQ: 45.6 -> 46.5 | Helped |
Closest papers by meaning
- Masked-Attention Mask Transformer for Universal Image Segmentation (CVPR 2022 · similarity 0.92)
- MaskLab: Instance Segmentation by Refining Object Detection With Semantic and Direction Features (CVPR 2018 · similarity 0.90)
- PEM: Prototype-based Efficient MaskFormer for Image Segmentation (CVPR 2024 · similarity 0.88)
- Mask-Adapter: The Devil is in the Masks for Open-Vocabulary Segmentation (CVPR 2025 · similarity 0.87)
- MaskDiff: Modeling Mask Distribution with Diffusion Probabilistic Model for Few-Shot Instance Segmentation (AAAI 2024 · similarity 0.87)
- MaSS13K: A Matting-level Semantic Segmentation Benchmark (CVPR 2025 · 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.