Research map / Detection and segmentation / CNN features
CrackFormer: Transformer Network for Fine-Grained Crack Detection
Huajun Liu, Xiangyu Miao, Christoph Mertz, Chengzhong Xu, Hui Kong
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 |
|---|---|---|---|
| replace 3x3 conv with Self-AB in encoder and decoder (ablation) | SegNet · CrackLS315 | ODS: gain on ODS and OIS is 9.8% and 8.9% | Improved |
| add Scal-AB (ablation) | SegNet · CrackLS315 | ODS: 9.7% ODS gain and an 8.9% OIS gain | Improved |
| apply Self-AB (ablation) | DeepCrack · CrackLS315 | ODS: 1.1% and 0.3% | Improved |
| apply Scal-AB (ablation) | DeepCrack · CrackLS315 | ODS: 0.9% and 0.5% | Improved |
| combine Self-AB and Scal-AB (ablation) | CrackFormer · CrackLS315 | ODS: 0.7% -0.9% | Improved |
| fuse all stages instead of finest stage S5 (ablation) | CrackFormer · CrackTree260 | ODS: 0.799 -> 0.881 | Improved |
| incremental multi-scale fusion S1+S2+S3+S4 vs single scale S1 (ablation) | CrackFormer · CrackTree260 | ODS: 0.680 -> 0.854 | Improved |
| fuse all scales vs finest scale S5 (average over datasets) (ablation) | CrackFormer · three crack datasets | ODS: 8.7% in average | Improved |
| use deeper stage output S5 vs S1 (ablation) | CrackFormer · Stone331 | ODS: 0.760 -> 0.809 | Improved |
| CrackFormer vs DeepCrack architecture | CrackFormer · CrackTree260 | ODS: 0.852 -> 0.881 | Improved |
| CrackFormer vs DeepCrack architecture | CrackFormer · CrackLS315 | ODS: 0.853 -> 0.871 | Improved |
| CrackFormer vs DeepCrack architecture | CrackFormer · Stone331 | ODS: 0.856 -> 0.877 | Improved |
Later papers that compared with this method
1 later accepted paper reports Crackformer as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Mind Marginal Non-Crack Regions: Clustering-Inspired Representation Learning for Crack Segmentation (CVPR 2024 · better on 1 of 1 shared result)
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 3 of 3 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
- Cracks in the Foundation: A Civil Infrastructure Dataset to Challenge Vision Foundation Models (NeurIPS 2026 · similarity 0.84)
- SCSegamba: Lightweight Structure-Aware Vision Mamba for Crack Segmentation in Structures (CVPR 2025 · similarity 0.83)
- Mind Marginal Non-Crack Regions: Clustering-Inspired Representation Learning for Crack Segmentation (CVPR 2024 · similarity 0.83)
- Wavelet and Prototype Augmented Query-based Transformer for Pixel-level Surface Defect Detection (CVPR 2025 · similarity 0.81)
- Attack-inspired Calibration Loss for Calibrating Crack Recognition (AAAI 2025 · similarity 0.81)
- TGBFormer: Transformer-GraphFormer Blender Network for Video Object Detection (AAAI 2025 · similarity 0.81)
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.
