Research map / Detection and segmentation / CNN features
ExFuse: Enhancing Feature Fusion for Semantic Segmentation
Zhenli Zhang, Xiangyu Zhang, Chao Peng, Xiangyang Xue, Jian Sun
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 semantic supervision (ablation) | GCN baseline (ResNeXt 101) · PASCAL VOC 2012 val | mIoU: 76.0 -> 77.5 | Improved |
| add layer rearrangement {8,8,9,8} (ablation) | GCN + SS · PASCAL VOC 2012 val | mIoU: 77.5 -> 78.3 | Improved |
| add semantic embedding branch (ablation) | GCN + SS + LR · PASCAL VOC 2012 val | mIoU: 78.3 -> 79.0 | Improved |
| add explicit channel resolution embedding (ablation) | GCN + SS + LR · PASCAL VOC 2012 val | mIoU: 78.3 -> 78.8 | Improved |
| add densely adjacent prediction (ablation) | GCN + SS + LR + SEB · PASCAL VOC 2012 val | mIoU: 79.0 -> 79.6 | Improved |
| add more output channels without DAP (ablation) | GCN + SS + LR + SEB · PASCAL VOC 2012 val | mIoU: 79.0 -> 79.1 | No clear effect |
| add auxiliary segmentation loss on deconvolution output (ablation) | GCN + SS + LR · PASCAL VOC 2012 val | mIoU: 78.3 -> 78.2 | No clear effect |
| replace deconvolution with sub-pixel upsample without supervision (ablation) | GCN + SS + LR · PASCAL VOC 2012 val | mIoU: 78.3 -> 77.6 | Hurt |
| add semantic supervision (auxiliary losses in pretraining) (ablation) | ResNet 50 · PASCAL VOC 2012 val | mIoU: 72.4 -> 73.5 | Improved |
| add semantic supervision (auxiliary losses in pretraining) (ablation) | ResNet 50 · ImageNet val | top-1 classification error: 24.15 -> 24.77 | Hurt |
| fuse low-level feature levels 1 and 2 into GCN (ablation) | original GCN ResNeXt 101 · PASCAL VOC 2012 val | mIoU: 75.98 -> 76.02 | No clear effect |
| fuse low-level feature levels 1 and 2 into ExFuse (ablation) | ExFuse · PASCAL VOC 2012 val | mIoU: 78.69 -> 80.04 | Improved |
Later papers that compared with this method
2 later accepted papers report ExFuse as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- Learnable Tree Filter for Structure-preserving Feature Transform (NeurIPS 2019 · better on 1 of 1 shared result)
- Expectation-Maximization Attention Networks for Semantic Segmentation (ICCV 2019 · better on 1 of 1 shared result)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Towards Bridging Semantic Gap to Improve Semantic Segmentation (ICCV 2019 · similarity 0.89)
- Gated Fully Fusion for Semantic Segmentation (AAAI 2020 · similarity 0.88)
- SED: A Simple Encoder-Decoder for Open-Vocabulary Semantic Segmentation (CVPR 2024 · similarity 0.87)
- Integrating Low-Level Visual Cues for Enhanced Unsupervised Semantic Segmentation (AAAI 2025 · similarity 0.87)
- Rethinking Decoders for Transformer-based Semantic Segmentation: A Compression Perspective (NeurIPS 2024 · similarity 0.87)
- Unveiling the Knowledge of CLIP for Training-Free Open-Vocabulary Semantic Segmentation (AAAI 2025 · similarity 0.86)
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.
