Research map / Detection and segmentation / CNN features
Learning Fully Dense Neural Networks for Image Semantic Segmentation
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 no feature reuse with dense feature reuse (ablation) | Encoder-decoder segmentation model · PASCAL VOC 2012 validation, stride 32 | mIoU: 77.2 -> 78.9 | Helped |
| Replace naive feature reuse with dense feature reuse (ablation) | Encoder-decoder segmentation model · PASCAL VOC 2012 validation, stride 32 | mIoU: 78.5 -> 78.9 | Helped |
| Use encoder stride 16 instead of stride 32 (ablation) | Encoder-decoder segmentation model · PASCAL VOC 2012 validation | mIoU: 78.9 -> 79.4 | Helped |
| Add boundary-focused pixel weighting with alpha (8,6,4,2,1) (ablation) | FDNet-16s · PASCAL VOC 2012 validation | mIoU: 79.4 -> 80.3 | Helped |
| Use exponential rather than poly attention weighting (ablation) | Boundary-aware loss on FDNet-16s · PASCAL VOC 2012 validation | mIoU: 80.0 vs 80.3; 80.9 best with exp | Helped |
| Increase poly attention lambda to 5 (ablation) | Boundary-aware loss on FDNet-16s · PASCAL VOC 2012 validation | mIoU: 80.3 -> 77.7 | Hurt |
| Use exponential attention weighting with lambda=0.75 (ablation) | Boundary-aware loss on FDNet-16s · PASCAL VOC 2012 validation | mIoU: 79.4 -> 80.9 | Helped |
| Fine-tune on the standard PASCAL VOC training images after trainaug training (ablation) | FDNet-16s · PASCAL VOC 2012 validation | mIoU: 82.1 -> 84.1 | Helped |
| Use multi-scale inference (ablation) | FDNet-16s · PASCAL VOC 2012 validation | mIoU: 80.9 -> 82.1 | Helped |
| Use FDNet-16s instead of RefineNet | RGB-only semantic segmentation · NYUDv2 | mIoU: 46.5 -> 47.4 | Helped |
Closest papers by meaning
- Fully Convolutional Neural Networks with Full-Scale-Features for Semantic Segmentation (AAAI 2017 · similarity 0.90)
- Semantic Segmentation With Boundary Neural Fields (CVPR 2016 · similarity 0.89)
- Dense Decoder Shortcut Connections for Single-Pass Semantic Segmentation (CVPR 2018 · similarity 0.89)
- Fast Neural Architecture Search of Compact Semantic Segmentation Models via Auxiliary Cells (CVPR 2019 · similarity 0.89)
- Knowledge Adaptation for Efficient Semantic Segmentation (CVPR 2019 · similarity 0.88)
- Unsupervised Semantic Segmentation Through Depth-Guided Feature Correlation and Sampling (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.