Living with Robots Elevator Button Dataset: 400 Images for Robot Navigation

400 JPEG images of elevator panels, with 292 taken across 24 buildings on the University of Texas at Austin campus and 108 sourced from the internet under varying conditions. The dataset includes JSON files with pixel-level annotations for buttons and labels, plus PyTorch model weights for segmentation and text recognition. It was created by Nicholas Verzic and last updated in October 2025.

Use Cases

• Train instance segmentation models to detect elevator buttons and labels based on the provided pixel-level annotations.
• Train scene text recognition models to read special characters on elevator panel labels.
• Develop models for associating buttons with their corresponding labels using the provided action-based labeling.
• Evaluate the robustness of vision systems under varying lighting, blur, and camera perspective conditions described in the dataset.

Strengths

• 400 annotated images provide a substantial base for model training.
• Includes 292 real-world images from 25 different elevators across 24 buildings, offering environmental diversity.
• Provides pre-trained model weights (segmentation and text recognition) to accelerate research and benchmarking.

Limitations

• Row count and file formats for the core data files are unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.
• Data may reflect geographic and architectural bias inherent to its primary collection from a single university campus.

Provenance

Source

Texas Data Repository Harvested Dataverse, authored by Nicholas Verzic.

Collection Method

Images collected via two methods: photographs taken on the University of Texas at Austin campus and images sourced from the internet.

Time Range

null

Freshness

Last updated 2025-10-15 07:01:42; freshness should be verified.

Geography

Primarily the University of Texas at Austin campus, with supplementary internet-sourced images of unspecified origin.