Research map / Detection and segmentation / CNN features
Cost Aggregation with 4D Convolutional Swin Transformer for Few-Shot Segmentation
Sunghwan Hong, Seokju Cho, Jisu Nam, Stephen Lin, Seungryong Kim
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| add Swin Transformer for cost aggregation to the baseline (ablation) | VAT baseline · FSS-1000 1-shot | mIoU: 80.0 -> 85.4 | Improved |
| replace non-overlapping VEM with overlapping-convolution VCM (ablation) | VAT · FSS-1000 1-shot | mIoU: 85.4 -> 87.0 | Improved |
| add guided pyramidal processing (GPP) (ablation) | VAT · FSS-1000 1-shot | mIoU: 87.0 -> 87.3 | Improved |
| add affinity-aware transformer decoder (ATD) (ablation) | VAT · FSS-1000 1-shot | mBA: 60.7 -> 68.0 | Improved |
| use transformers in the decoder instead of convolutions only (ablation) | ATD · FSS-1000 1-shot | mIoU: 87.3 -> 90.3 | Improved |
| use 4D Conv. Swin transformer instead of center-pivot 4D convolutions as aggregator (ablation) | VAT · FSS-1000 1-shot | mIoU: 88.1 -> 90.3 | Improved |
| use Swin aggregator instead of Linear transformer (ablation) | VAT · FSS-1000 1-shot | mIoU: 87.7 -> 90.3 | Improved |
| use VCM instead of VEM with Swin Transformer (ablation) | VAT · FSS-1000 1-shot | mIoU: 89.9 -> 90.3 | Improved |
| use VCM instead of VEM with Linear Transformer (ablation) | Linear Transformer aggregator · FSS-1000 1-shot | mIoU: 87.0 -> 87.7 | Improved |
| swap ResNet101 backbone for Swin transformer backbone (ablation) | VAT · FSS-1000 1-shot | mIoU: 90.3 -> 89.8 | No clear effect |
| add PFE-Net style data augmentation (ablation) | VAT · PASCAL-5i mean mIoU | mIoU: 67.9 -> 67.3 | Hurt |
| add CATs data augmentation (ablation) | VAT · PASCAL-5i mean mIoU | mIoU: 67.9 -> 66.1 | Hurt |
Closest papers by meaning
- Container: Context Aggregation Networks (NeurIPS 2021 · similarity 0.87)
- CATs: Cost Aggregation Transformers for Visual Correspondence (NeurIPS 2021 · similarity 0.87)
- Doubly Deformable Aggregation of Covariance Matrices for Few-Shot Segmentation (ECCV 2022 · similarity 0.87)
- Unlocking the Potential of Pre-trained Vision Transformers for Few-Shot Semantic Segmentation through Relationship Descriptors (CVPR 2024 · similarity 0.85)
- Unifying Feature and Cost Aggregation with Transformers for Semantic and Visual Correspondence (ICLR 2024 · similarity 0.85)
- Rethinking Decoders for Transformer-based Semantic Segmentation: A Compression Perspective (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.
