Research map / Detection and segmentation / CNN features
Large Kernel Matters -- Improve Semantic Segmentation by Global Convolutional Network
Chao Peng, Xiangyu Zhang, Gang Yu, Guiming Luo, Jian Sun
Check 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 |
|---|---|---|---|
| Increase GCN kernel size from 3 to 15 (ablation) | GCN segmentation model · PASCAL VOC 2012 val | mean IoU: 70.1 -> 74.5 | Improved |
| Replace GCN with a trivial k×k convolution (ablation) | GCN segmentation model · PASCAL VOC 2012 val | mean IoU: 70.1 -> 69.8; 71.1 -> 70.4; 72.8 -> 69.6; 73.4 -> 68.8 | Hurt |
| Replace GCN with a stack of small convolutions (ablation) | GCN segmentation model · PASCAL VOC 2012 val, large equivalent kernels | mean IoU: GCN/Stack: 70.1/69.8, 71.1/71.8, 72.8/71.3, 73.4/69.5, 73.7/67.5 | Mixed |
| Reduce intermediate channels in a convolution stack (ablation) | Stack of small convolutions, compared with GCN · PASCAL VOC 2012 val | mean IoU: 71.3 -> 70.4 -> 68.8; GCN: 72.8 | Hurt |
| Add boundary refinement to GCN (ablation) | GCN segmentation model · PASCAL VOC 2012 val, boundary pixels | accuracy: 71.5 -> 73.4 | Improved |
| Add GCN with k=15 (ablation) | 1×1 convolution baseline · PASCAL VOC 2012 val, internal pixels | accuracy: 93.9 -> 95.0 | Improved |
| Add boundary refinement to GCN (ablation) | GCN segmentation model · PASCAL VOC 2012 val, internal pixels | accuracy: 95.0 -> 95.1 | No clear effect |
| Replace the first two ResNet bottleneck layers with GCN (ablation) | ResNet50 segmentation model · PASCAL VOC 2012 val | mean IoU: 65.7 -> 71.2 | Improved |
| Use ResNet50-GCN instead of ResNet50 as the pretrained model, with GCN + BR (ablation) | GCN + BR segmentation model · PASCAL VOC 2012 val | mean IoU: 72.3 -> 72.5 | No clear effect |
| Add boundary refinement to GCN (ablation) | PASCAL VOC 2012 Stage-3 model · PASCAL VOC 2012 val | mean IoU: 78.7 -> 80.3 | Improved |
| Apply multi-scale evaluation (ablation) | PASCAL VOC 2012 Stage-3 GCN + BR · PASCAL VOC 2012 val | mean IoU: 80.3 -> 80.4 | No clear effect |
| Apply denseCRF post-processing after multi-scale evaluation (ablation) | PASCAL VOC 2012 Stage-3 GCN + BR · PASCAL VOC 2012 val | mean IoU: 80.4 -> 81.0 | Improved |
Closest papers by meaning
- Scaling Semantic Segmentation Beyond 1K Classes on a Single GPU (ICCV 2021 · similarity 0.89)
- Revisiting Dilated Convolution: A Simple Approach for Weakly- and Semi-Supervised Semantic Segmentation (CVPR 2018 · similarity 0.89)
- DeepLab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFs (TPAMI 2017 · similarity 0.89)
- K-Net: Towards Unified Image Segmentation (NeurIPS 2021 · similarity 0.89)
- Rethinking Semantic Segmentation From a Sequence-to-Sequence Perspective With Transformers (CVPR 2021 · similarity 0.88)
- Fully Convolutional Networks for Semantic Segmentation (TPAMI 2016 · 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.
