Research map / Detection and segmentation / CNN features
EfficientFCN: Holistically-guided Decoding for Semantic Segmentation
Jianbo Liu, Junjun He, Jiawei Zhang, Jimmy S. Ren, Hongsheng Li
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 OS=16 and OS=32 features to the fused map used for codebook generation (ablation) | EfficientFCN · PASCAL Context | mIoU: 54.8 -> 55.1 | Improved |
| add OS=8 features as well to the fused map used for codebook generation (ablation) | EfficientFCN · PASCAL Context | mIoU: 55.1 -> 55.3 | Improved |
| add OS=16 features to the fused map used for codeword assembly coefficients (ablation) | EfficientFCN · PASCAL Context | mIoU: 47.9 -> 52.1 | Improved |
| add OS=32 features as well to the fused map used for codeword assembly coefficients (ablation) | EfficientFCN · PASCAL Context | mIoU: 52.1 -> 55.3 | Improved |
| add the average codeword base transfer to the assembly guidance features (ablation) | EfficientFCN · PASCAL Context | mIoU: 54.2 -> 55.3 | Improved |
| increase number of codewords from 32 to 512 (ablation) | EfficientFCN · PASCAL Context | mIoU: 54.5 -> 55.5 | Improved |
| increase number of codewords from 512 to 1024 (ablation) | EfficientFCN · PASCAL Context | mIoU: 55.5 -> 55.1 | Hurt |
| increase number of codewords from 256 to 512 (ablation) | EfficientFCN · PASCAL Context | GFlops: 69.6 -> 72.1 | Hurt |
| replace bilinear-upsampling U-Net decoder with holistically-guided decoder | ResNet101 encoder-decoder · PASCAL Context | mIoU: 49.3 -> 55.3 | Improved |
| replace dilated ResNet101 FCN with non-dilated ResNet101 plus holistically-guided decoder | dilatedFCN-8s · PASCAL Context | mIoU: 47.2 -> 55.3 | Improved |
| replace dilated ResNet101 FCN with non-dilated ResNet101 plus holistically-guided decoder | dilatedFCN-8s · PASCAL Context | GFlops: 223.6 -> 69.6 | Improved |
| add COCO pre-training | EfficientFCN · PASCAL VOC 2012 test | mIoU: 85.4 -> 87.6 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DANet — DANet: Divergent Activation for Weakly Supervised Object Localization (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
- A Holistically-Guided Decoder for Deep Representation Learning With Applications to Semantic Segmentation and Object Detection (TPAMI 2021 · similarity 0.95)
- Dense Decoder Shortcut Connections for Single-Pass Semantic Segmentation (CVPR 2018 · similarity 0.90)
- Learning Fully Dense Neural Networks for Image Semantic Segmentation (AAAI 2019 · similarity 0.90)
- Fully Convolutional Networks for Semantic Segmentation (TPAMI 2016 · similarity 0.89)
- SED++: A Simple Encoder-Decoder for Improved Open-Vocabulary Semantic Segmentation (TPAMI 2025 · similarity 0.89)
- SED: A Simple Encoder-Decoder for Open-Vocabulary Semantic Segmentation (CVPR 2024 · similarity 0.89)
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.
