Research map / Detection and segmentation / CNN features
ICNet for Real-Time Semantic Segmentation on High-Resolution Images
Hengshuang Zhao, Xiaojuan Qi, Xiaoyong Shen, Jianping Shi, Jiaya Jia
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 medium-resolution branch to low-resolution branch (sub4 to sub24) (ablation) | ICNet · Cityscapes val | mIoU: 59.6 -> 66.5 | Improved |
| add high-resolution branch (sub24 to sub124) (ablation) | ICNet · Cityscapes val | mIoU: 66.5 -> 67.7 | Improved |
| use full cascade ICNet instead of half-compressed PSPNet50 | PSPNet50 compressed to half · Cityscapes val | inference time (ms): 170 -> 33 | Improved |
| add cascade label guidance (ablation) | ICNet with CFF · Cityscapes val | mIoU: 66.8 -> 67.7 | Improved |
| use CFF instead of 3x3 deconvolution (ablation) | ICNet · Cityscapes val | mIoU: 66.7 -> 67.7 | Improved |
| use CFF instead of 5x5 deconvolution (ablation) | ICNet · Cityscapes val | mIoU: 66.7 -> 67.7 | Improved |
| use CFF instead of 7x7 deconvolution (ablation) | ICNet · Cityscapes val | inference time (ms): 38 -> 33 | Improved |
| use CFF instead of 7x7 deconvolution (ablation) | ICNet · Cityscapes val | mIoU: 68.0 -> 67.7 | No clear effect |
| increase feature downsampling factor from 8 to 32 | PSPNet50 · Cityscapes val | mIoU: 71.7 -> 67.1 mIoU, 446 -> 131 ms | Hurt |
| keep only 0.25 of kernels (model compression) | PSPNet50 · Cityscapes val | mIoU: 71.7 -> 59.4 mIoU, 446 -> 72 ms | Hurt |
| train with both fine and coarse data | ICNet · Cityscapes test | mIoU: 69.5 -> 70.6 | Improved |
| use ICNet instead of PSPNet50 on CamVid | PSPNet50 · CamVid test | mIoU: 69.1 -> 67.1 mIoU, 185 -> 36 ms | Mixed |
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.
- SegNet — SegNet: A Deep Convolutional Encoder-Decoder Architecture for Image Segmentation (TPAMI 2017) · better on 2 of 2 shared results
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
- CascadePSP: Toward Class-Agnostic and Very High-Resolution Segmentation via Global and Local Refinement (CVPR 2020 · similarity 0.89)
- On the Importance of Label Quality for 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)
- Not All Pixels Are Equal: Difficulty-Aware Semantic Segmentation via Deep Layer Cascade (CVPR 2017 · similarity 0.89)
- Semantic Projection Network for Zero- and Few-Label Semantic Segmentation (CVPR 2019 · similarity 0.88)
- SCTNet: Single-Branch CNN with Transformer Semantic Information for Real-Time Segmentation (AAAI 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.
