Research map / Detection and segmentation / CNN features
ISDNet: Integrating Shallow and Deep Networks for Efficient Ultra-High Resolution Segmentation
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 |
|---|---|---|---|
| Integrate shallow full-resolution processing with a deep downsampled-image branch (ablation) | ISDNet · DeepGlobe | mIoU: 61.40, 73.23, 56.39, 70.30 -> 71.69 | Helped |
| Use a deep branch with ¼-scale input instead of only the deep branch with full-scale input (ablation) | Deep branch · DeepGlobe | mIoU: 61.40 -> 73.23 | Helped |
| Use RAF rather than addition fusion (ablation) | ISDNet · DeepGlobe | mIoU: 72.20 -> 73.30 | Helped |
| Use RAF rather than concatenation fusion (ablation) | ISDNet · DeepGlobe | mIoU: 72.42 -> 73.30 | Helped |
| Use RAF rather than channel-wise attention fusion (ablation) | ISDNet · DeepGlobe | mIoU: 71.88 -> 73.30 | Helped |
| Add super-resolution loss (ablation) | ISDNet with baseline losses · DeepGlobe | mIoU: 72.31 -> 72.55 | Helped |
| Add structure distillation loss alongside super-resolution loss (ablation) | ISDNet · DeepGlobe | mIoU: 72.55 -> 72.70 | Helped |
| Replace shallow-branch RGB input with high-frequency residuals (ablation) | ISDNet · DeepGlobe | mIoU: 72.70 -> 73.30 | Helped |
| Use ISDNet instead of MagNet | DeepGlobe segmentation · DeepGlobe | mIoU: 72.96 -> 73.30 | Helped |
| Use ISDNet instead of PPN | DeepGlobe segmentation · DeepGlobe | FPS: 12.90 -> 27.70 | Helped |
| Use ISDNet instead of FCtL | Inria Aerial segmentation · Inria Aerial | FPS: 0.04 -> 6.90 | Helped |
| Integrate PSPNet using the ISD framework rather than standalone PSPNet | PSPNet · Cityscapes | FPS: 15.15 -> 58.29 | Helped |
Closest papers by meaning
- Patch Proposal Network for Fast Semantic Segmentation of High-Resolution Images (AAAI 2020 · similarity 0.88)
- Collaborative Global-Local Networks for Memory-Efficient Segmentation of Ultra-High Resolution Images (CVPR 2019 · similarity 0.87)
- PIDNet: A Real-Time Semantic Segmentation Network Inspired by PID Controllers (CVPR 2023 · similarity 0.87)
- Toward Real Ultra Image Segmentation: Leveraging Surrounding Context to Cultivate General Segmentation Model (NeurIPS 2024 · similarity 0.86)
- F2Net: A Frequency-Fused Network for Ultra-High Resolution Remote Sensing Segmentation (CVPR 2026 · similarity 0.86)
- Unsupervised Semantic Segmentation Through Depth-Guided Feature Correlation and Sampling (CVPR 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.