Research map / Detection and segmentation / CNN features
SegViT: Semantic Segmentation with Plain Vision Transformers
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 SETR with ATM while retaining cross-entropy loss (ablation) | ViT-Base segmentation on ADE20K · ADE20K | mIoU (ss): 46.5 -> 47.0 | Helped |
| Use mask and classification supervision with ATM (ablation) | ViT-Base segmentation on ADE20K · ADE20K | mIoU (ss): 46.5 -> 49.6 | Helped |
| Add layer 6 features to layer 12 (ablation) | SegViT · ADE20K | mIoU (ss): 49.6 -> 50.9 | Helped |
| Use layers 6, 8, and 12 (ablation) | SegViT · ADE20K | mIoU (ss): 49.6 -> 51.3 | Helped |
| Add layer 3, producing four input layers (ablation) | SegViT · ADE20K | mIoU (ss): 51.3 -> 51.2 | Hurt |
| Down-sample Q with convolution rather than nearest sampling (ablation) | QD · ADE20K | mIoU (ss): 53.9 -> 44.5 | Hurt |
| Down-sample Q, K, and V instead of Q only (ablation) | QD · ADE20K | mIoU (ss): 53.9 -> 52.6 | Hurt |
| Apply QD without adding QU (ablation) | Shrunk SegViT · ADE20K | mIoU (ss): 49.6 -> 46.9 | Hurt |
| Add QU to the QD structure (ablation) | Shrunk SegViT · ADE20K | mIoU (ss): 46.9 -> 50.0 | Helped |
| Use token merge rather than FPN for multi-level ViT features (ablation) | ViT-Base segmentation · ADE20K | mIoU (ss): 46.7 -> 50.6 | Helped |
| Apply attention-mask loss rather than dot-product loss with token merge (ablation) | ViT-Base segmentation · ADE20K | mIoU (ss): 50.6 -> 51.2 | Helped |
| Use the Shrunk structure instead of full SegViT | SegViT with ViT-Large · ADE20K val | GFLOPs: 637.9 -> 373.5 | Helped |
Closest papers by meaning
- Rethinking Decoders for Transformer-based Semantic Segmentation: A Compression Perspective (NeurIPS 2024 · similarity 0.88)
- Pixels to Tokens: Token Space Efficient Active Learning for Low-Budget Semantic Segmentation (NeurIPS 2026 · similarity 0.88)
- MaskViM: Domain Generalized Semantic Segmentation with State Space Models (AAAI 2025 · similarity 0.88)
- Your ViT is Secretly an Image Segmentation Model (CVPR 2025 · similarity 0.88)
- Seg-ReSearch: Segmentation with Interleaved Reasoning and External Search (ICML 2026 · similarity 0.88)
- SED: A Simple Encoder-Decoder for Open-Vocabulary Semantic Segmentation (CVPR 2024 · 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.