Research map / Architectures and efficiency / RNNs and normalization
In Defense of Pre-Trained ImageNet Architectures for Real-Time Semantic Segmentation of Road-Driving Images
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use ImageNet pre-training instead of random initialization | SwiftNet single-scale models · Cityscapes | mIoU: 5 mIoU percentage points | Helped |
| Use ImageNet pre-training instead of random initialization | SwiftNet models · CamVid | mIoU: 7-9pp | Helped |
| Add interleaved pyramid fusion (ablation) | SwiftNet with ResNet-18 · CamVid test | mIoU: 72.58 -> 73.86 | Helped |
| Add interleaved pyramid fusion (ablation) | SwiftNet with MobileNet V2 · CamVid test | mIoU: 71.56 -> 73.08 | Helped |
| Add interleaved pyramid fusion (ablation) | SwiftNet · Cityscapes | mIoU | No clear effect |
| Add lateral connections between encoder and decoder (ablation) | SwiftNet single-scale model · Cityscapes val | mIoU: 72.93 -> 75.35 | Helped |
| Use 128 rather than 64 upsampling features (ablation) | SwiftNetRN-18 · Cityscapes val | mIoU: 69.50 -> 70.35 | Helped |
| Increase upsampling features beyond 128 (ablation) | SwiftNetRN-18 · Cityscapes val | mIoU: 70.35 -> 70.26; 70.35 -> 70.63 | No clear effect |
| Add SPP (ablation) | ResNet-18 · Cityscapes val | effective receptive field horizontal: 84.47 -> 154.07 | Helped |
| Use interleaved pyramid fusion (ablation) | ResNet-18 · Cityscapes val | effective receptive field horizontal: 84.47 -> 185.22 | Helped |
| Take lateral features after the following ReLU rather than after the residual-block sum (ablation) | SwiftNet upsampling decoder · Cityscapes validation | mIoU | Hurt |
| Replace the 3×3 decoder convolution with a 1×1 convolution (ablation) | SwiftNet upsampling decoder · Cityscapes validation | mIoU | Hurt |
Closest papers by meaning
- Full-Resolution Residual Networks for Semantic Segmentation in Street Scenes (CVPR 2017 · similarity 0.90)
- DenseASPP for Semantic Segmentation in Street Scenes (CVPR 2018 · similarity 0.89)
- Weakly Supervised Semantic Segmentation for Driving Scenes (AAAI 2024 · similarity 0.88)
- RoadSceneBench: A Lightweight Benchmark for Mid-Level Road Scene Understanding (CVPR 2026 · similarity 0.86)
- GSENet:Global Semantic Enhancement Network for Lane Detection (AAAI 2024 · similarity 0.86)
- LaneSegNet: Map Learning with Lane Segment Perception for Autonomous Driving (ICLR 2024 · 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.