Research map / Detection and segmentation / CNN features
Exploring Context with Deep Structured Models for Semantic Segmentation
Guosheng Lin, Chunhua Shen, Anton van den Hengel, Ian D. Reid
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 sliding pyramid pooling (ablation) | FullyConvNet baseline · NYUDv2 (40 classes) | IoU: 30.5 -> 32.4 | Improved |
| add multi-scale inputs (ablation) | FullyConvNet + pyramid pooling · NYUDv2 (40 classes) | IoU: 32.4 -> 37.0 | Improved |
| add boundary refinement (ablation) | FullyConvNet + multi-scales · NYUDv2 (40 classes) | IoU: 37.0 -> 38.3 | Improved |
| add CNN contextual pairwise potentials (ablation) | FCN with refinement · NYUDv2 (40 classes) | IoU: 38.3 -> 40.6 | Improved |
| use 1 pairwise + 1 unary CRF instead of 4-unary ensemble (ablation) | unary-only ensemble · NYUDv2 (40 classes) | IoU: 38.7 -> 38.9 | Improved |
| add a second unary to the contextual CRF (ablation) | 1 pairwise + 1 unary · NYUDv2 (40 classes) | IoU: 38.9 -> 39.2 | Improved |
| ensemble more unary-only networks (1 to 4) (ablation) | unary network · NYUDv2 (40 classes) | IoU: 37.0 -> 38.7 | Improved |
| add COCO pretraining/training data | ours · PASCAL VOC 2012 test | mean IoU: 75.3 -> 77.8 | Improved |
| use contextual CRF model instead of FullyConvNet + refine | FullyConvNet + refine · COCO (80 classes) | IoU: 41.3 -> 46.8 | Improved |
| use RGB-only contextual CRF instead of FCN-HHA using RGB-D | FCN-HHA · NYUDv2 (40 classes) | IoU: 34.0 -> 40.6 | Improved |
| add COCO pretraining | ours · KITTI | IoU: 68.5 -> 70.3 | Improved |
| add refinement to fully convolutional network | FullyConvNet · COCO (80 classes) | mean accuracy: 56.9 -> 55.0 | Hurt |
Closest papers by meaning
- Efficient Piecewise Training of Deep Structured Models for Semantic Segmentation (CVPR 2016 · similarity 0.97)
- Semantic Segmentation via Structured Patch Prediction, Context CRF and Guidance CRF (CVPR 2017 · similarity 0.92)
- Semantic Segmentation With Boundary Neural Fields (CVPR 2016 · similarity 0.90)
- Fast, Exact and Multi-scale Inference for Semantic Image Segmentation with Deep Gaussian CRFs (ECCV 2016 · similarity 0.90)
- Hierarchically Gated Deep Networks for Semantic Segmentation (CVPR 2016 · similarity 0.89)
- Unsupervised Semantic Segmentation Through Depth-Guided Feature Correlation and Sampling (CVPR 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.
