ClimbInst: Climbing Hold Instance Segmentation Dataset
ClimbInst is an instance segmentation dataset for climbing holds and volumes on indoor climbing walls. It was developed to support research in climbing-hold instance segmentation and related computer vision applications.
The dataset contains 1,509 images with a total of 89,020 annotated instances. Climbing holds and volumes are represented using a single unified class.
Dataset Structure
The dataset is divided into training, validation, and test splits:
ClimbInst/
├── train/
│ ├── images/
│ └── annotations/
├── validation/
│ ├── images/
│ └── annotations/
└── test/
├── images/
└── annotations/
Dataset Statistics
| Split | Images | Annotated Instances |
|---|---|---|
| Train | 1,169 | 71,319 |
| Validation | 290 | 14,268 |
| Test | 50 | 3,433 |
| Total | 1,509 | 89,020 |
The test set is independent of the training and validation data.
Dataset Construction
Images used to construct ClimbInst were collected from multiple sources, including publicly available climbing-wall datasets and separately collected climbing-wall imagery.
Initial instance masks were generated using SAM3 as an annotation-assistance mechanism. The generated masks were subsequently manually inspected and corrected using Labelme. Incorrect masks were refined or removed, and climbing holds missed during the initial annotation process were manually added.
Both climbing holds and climbing volumes are annotated as a single unified class.
Annotations
The dataset provides instance-level segmentation annotations for climbing holds and volumes.
Each training, validation, and test split contains its corresponding images and annotations.
Intended Use
ClimbInst is intended to support research and development involving:
Climbing-hold instance segmentation
Climbing-hold detection
Computer vision for indoor climbing
Digital representation of climbing walls
Climbing route selection and analysis
Related computer vision applications
Dataset Sources
ClimbInst incorporates climbing-wall imagery collected from multiple sources, including publicly available datasets and separately collected images.
The publicly available datasets used during dataset construction include:
Indoor Climbing Gym Hold and Route Segmentation — T. Sláma, Kaggle
https://www.kaggle.com/datasets/tomasslama/indoor-climbing-gym-hold-segmentation
Climbing Holds and Volumes — Roboflow
https://app.roboflow.com/weedlabelling/climbing-holds-and-volumes-zdgn3/browse
License
License information is provided according to the applicable usage and redistribution terms of the constituent image sources.