Nordic Skiing Video Dataset
This site provides access to a multi-view video dataset of cross-country skiing technique, along with pretrained AlphaPose model weights, annotated data, and labeled gear cycles for research use.
About
This dataset contains drone-captured video recordings of 12 skiers performing the skating technique in gears 2, 3, and 4. The skiers range from beginners to national-level athletes, providing a wide variation in technique quality and style. Several high-level skiers also performed common technical mistakes, making the dataset suitable for research on error detection.
The current version includes approximately 155 videos, primarily filmed from side and front perspectives. All videos were recorded at Ormberget in Luleå, northern Sweden, with a resolution of 1920×1080 pixels and a frame rate of 30 frames per second.
A subset of the videos has been annotated using the Computer Vision Annotation Tool (CVAT) in the Halpe26 format. In addition, AlphaPose model weights, raw video recordings, annotation JSON files, and labeled cycles suitable for gear classification are made available on this site.
Data Downloads
Model Weights
- RegLoss100EpochNoFlipDPG.pth – Finetuned AlphaPose weights. Download
Raw Video Dataset
AlphaPose Finetuning Dataset
- Videos used for annotations – Edited videos used for annotating ground truth data for joints. Download
- Annotations in JSON files – Ground truth labels of joint positions. Download
- Test data videos – Short sequences used for testing AlphaPose predicitons. Download
- Test data JSON annotations – Ground truth test data for AlphaPose. Download
Annotated Gear Cycles Dataset
- Labeled cycles – Gear cycles labaled as gear 2, gear 3, gear 4, or unknown for each video. Download
Code Repository
Contains code for dataset preprocessing, model training, and examples for researchers.
Go to GitHubPublication
[Add citation / DOI / BibTeX here]
Contact Us
[Your email or lab contact info]