TRails: Top Rail Segmentation
Authors/Creators
Contributors
Annotator (2):
Contact person (2):
Data collector (3):
Project member:
Work package leader:
et al. Show all 11 contributors
Annotator (2):
Contact person (2):
Data collector (3):
Project member:
Work package leader:
Description
Description
1280×720 resolution) on the LEONARDO platform (https://struktonrail.com/innovations/leonardo-inspection-train/). TRails specifically targets structural track complexity, varying ballast conditions, and safety-critical switch assemblies.- Regular Rails — 1026 images
- Hard Examples (corner cases of regular rails) — 12 images
- Switch Rails — 223 images
- Frogs — 96 images
Data Format & Organization:
- Images/: High-resolution raw RGB images in .jpg format (1280 x 720).
- Labels/Masks/: Pixel-level ground-truth semantic segmentation masks in .png format.
- Labels/JSONs/: Object-level bounding annotations and metadata in .json format.
Primary Applications:
This benchmark is curated to advance intelligent railway perception, including:
-
Thin-structure semantic segmentation and boundary delineation
-
Track gauge metrology and switch-blade gap verification
-
Region of Interest (ROI) isolation for downstream unsupervised defect detection
-
High-definition railway digital twin and map generation
Acknowledgements
Funded by the European Union. Views and opinion expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or Europe’s Rail Joint Undertaking. Neither the European Union nor the granting authority can be held responsible for them. The project FP3-IAM4Rail (Grant Agreement No. 101101966) is supported by the Europe's Rail Joint Undertaking and its members.
License
The license is specified in the License.txt file.
Files
TRails_Dataset.zip
Additional details
Identifiers
Funding
Dates
- Submitted
-
2026-09-25Image dataset containing semantic lables of rails from top-view.