Research map / Detection and segmentation / CNN features
Recurrent U-Net for Resource-Constrained Segmentation
Wei Wang, Kaicheng Yu, Joachim Hugonot, Pascal Fua, Mathieu Salzmann
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 segmentation mask as recurrent input (ablation) | DRU(4) · EYTH validation | mIoU: 0.826 -> 0.836 | Improved |
| use DRU gated recurrent unit instead of plain recurrent unit at bottleneck (Rec-Mid) (ablation) | recurrent U-Net · EYTH validation | mIoU: 0.814 -> 0.826 | Improved |
| replace U-Net by recurrent DRU(4) on Massachusetts Roads | U-Net-G · Massachusetts Roads | mIoU: 0.479 -> 0.560 | Improved |
| add recurrence to VGG16 U-Net | U-Net-VGG16 · Massachusetts Roads | mIoU: 0.521 -> 0.571 | Improved |
| add recurrent DRU(4) instead of Rec-Simple (recursion on mask with topology-aware loss removed) (ablation) | U-Net backbone · Massachusetts Roads | mIoU: 0.534 -> 0.560 | Improved |
| double U-Net channels (ablation) | DRU(4) · Cityscapes val | mIoU: 0.532 -> 0.627 | Improved |
| double U-Net channels (ablation) | U-Net-G · Cityscapes val | mIoU: 0.429 -> 0.476 | Improved |
| add recurrence DRU(4) to U-Net-G | U-Net-G · Cityscapes val | mIoU: 0.429 -> 0.532 | Improved |
| use pretrained VGG16 encoder with DRU (ablation) | DRU(4) · KBH hand segmentation | mIoU: 0.935 -> 0.954 | Improved |
| use pretrained VGG16 encoder with DRU (ablation) | DRU(4) · DRIVE retina vessel | mIoU: 0.821 -> 0.817 | Hurt |
| use DRU(4) instead of pretrained RefineNet | RefineNet · HOF hand segmentation | mIoU: 0.766 -> 0.709 | Hurt |
| use DRU(4) instead of RefineNet | RefineNet · EYTH hand segmentation | mIoU: 0.688 -> 0.849 | Improved |
Closest papers by meaning
- End-To-End Instance Segmentation With Recurrent Attention (CVPR 2017 · similarity 0.86)
- Non-Local U-Nets for Biomedical Image Segmentation (AAAI 2020 · similarity 0.86)
- Toward Real Ultra Image Segmentation: Leveraging Surrounding Context to Cultivate General Segmentation Model (NeurIPS 2024 · similarity 0.86)
- Upping the Game: How 2D U-Net Skip Connections Flip 3D Segmentation (NeurIPS 2024 · similarity 0.84)
- Rethinking U-Net: Task-Adaptive Mixture of Skip Connections for Enhanced Medical Image Segmentation (AAAI 2025 · similarity 0.84)
- U-KAN Makes Strong Backbone for Medical Image Segmentation and Generation (AAAI 2025 · similarity 0.84)
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.
