Research map / Detection and segmentation / CNN features
FasterSeg: Searching for Faster Real-time Semantic Segmentation
Read the paperCheck 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 |
|---|---|---|---|
| add a second, different-resolution branch (ablation) | FasterSeg search space with fixed expansion ratio 8 · Cityscapes val | mIoU (%): 66.9 -> 69.5 | Helped |
| add a second, different-resolution branch (ablation) | FasterSeg search space with fixed expansion ratio 8 · Cityscapes val | FPS: 177.8 -> 119.9 | Hurt |
| search expansion ratios to eliminate redundant channels (ablation) | two-branch FasterSeg student · Cityscapes val | FPS: 119.9 -> 163.9 | Helped |
| search expansion ratios to eliminate redundant channels (ablation) | two-branch FasterSeg student · Cityscapes val | mIoU (%): 69.5 -> 70.5 | Helped |
| train the co-searched student with teacher knowledge distillation (ablation) | co-searched FasterSeg student · Cityscapes val | mIoU (%): 70.5 -> 73.1 | Helped |
| use co-searching rather than pruning the teacher to make the student (ablation) | teacher-student compression · Cityscapes val | mIoU (%): 66.1 -> 73.1 | Helped |
| use co-searching rather than pruning the teacher to make the student (ablation) | teacher-student compression · Cityscapes val | FPS: 146.7 -> 163.9 | Helped |
| use zoomed convolution instead of standard convolution (ablation) | operator candidates · 1080Ti TensorRT operator benchmark | Latency (ms): 0.15 -> 0.09 | Helped |
| use dilated convolution instead of standard convolution (ablation) | operator candidates · 1080Ti TensorRT operator benchmark | Latency (ms): 0.15 -> 0.25 | Hurt |
| use group-2 convolution instead of standard convolution (ablation) | operator candidates · 1080Ti TensorRT operator benchmark | Latency (ms): 0.15 -> 0.13 | Helped |
| apply latency regularization separately across operators, resolutions, and expansion ratios (ablation) | FasterSeg architecture search · Cityscapes validation | mIoU (%) | Helped |
Closest papers by meaning
- HyperSeg: Patch-Wise Hypernetwork for Real-Time Semantic Segmentation (CVPR 2021 · similarity 0.92)
- Seg-ReSearch: Segmentation with Interleaved Reasoning and External Search (ICML 2026 · similarity 0.88)
- SasWOT: Real-Time Semantic Segmentation Architecture Search WithOut Training (AAAI 2024 · similarity 0.88)
- APSeg: Auto-Prompt Network for Cross-Domain Few-Shot Semantic Segmentation (CVPR 2024 · similarity 0.87)
- Diffusion-Driven Two-Stage Active Learning for Low-Budget Semantic Segmentation (NeurIPS 2025 · similarity 0.87)
- Improving Semi-Supervised Semantic Segmentation with Sliced-Wasserstein Feature Alignment and Uniformity (CVPR 2025 · 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.