Research map / Detection and segmentation / CNN features
Rethinking Semantic Segmentation From a Sequence-to-Sequence Perspective With Transformers
Check 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 the FCN encoder with a pure transformer sequence-to-sequence encoder | Semantic segmentation · ADE20K val, single-scale | mIoU: 42.17 -> 48.64 | Helped |
| Use progressive upsampling instead of one-shot upsampling (ablation) | SETR · Cityscapes val, 80k, single-scale | mIoU: 77.90 -> 79.34 | Helped |
| Use multi-level feature aggregation instead of progressive upsampling (ablation) | SETR · ADE20K val, single-scale | mIoU: 48.58 -> 48.64 | Helped |
| Use multi-level feature aggregation instead of progressive upsampling (ablation) | SETR · Cityscapes val, 80k, single-scale | mIoU: 79.34 -> 77.24 | Hurt |
| Replace one-shot upsampling with progressive upsampling (ablation) | SETR · Pascal Context val, 80k, single-scale | mIoU: 52.89 -> 54.40 | Helped |
| Use T-Large instead of T-Base (ablation) | SETR-PUP · Cityscapes val, 80k | mIoU: 78.02 -> 79.34 | Helped |
| Use ImageNet pre-training instead of random initialization (ablation) | SETR-PUP · Cityscapes val | mIoU: 42.27 -> 79.34 | Helped |
| Use multi-scale inference instead of single-scale inference (ablation) | SETR-MLA · ADE20K val | mIoU: 48.64 -> 50.28 | Helped |
| Use multi-scale inference instead of single-scale inference (ablation) | SETR-MLA · Pascal Context val | mIoU: 54.87 -> 55.83 | Helped |
| Train longer, 80k rather than 40k iterations (ablation) | SETR-PUP · Cityscapes val, single-scale | mIoU: 78.39 -> 79.34 | Helped |
| Use DeiT ImageNet-1K pre-training rather than ViT ImageNet-21K pre-training (ablation) | SETR-PUP · Cityscapes val, 80k | mIoU: 78.39 -> 79.45 | Helped |
Closest papers by meaning
- SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers (NeurIPS 2021 · similarity 0.90)
- Learning Content-Enhanced Mask Transformer for Domain Generalized Urban-Scene Segmentation (AAAI 2024 · similarity 0.90)
- Context Encoding for Semantic Segmentation (CVPR 2018 · similarity 0.90)
- FeedFormer: Revisiting Transformer Decoder for Efficient Semantic Segmentation (AAAI 2023 · similarity 0.89)
- Rethinking Decoders for Transformer-based Semantic Segmentation: A Compression Perspective (NeurIPS 2024 · similarity 0.89)
- SCTNet: Single-Branch CNN with Transformer Semantic Information for Real-Time Segmentation (AAAI 2024 · similarity 0.88)
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.