Sidewalk Images from Belgium with Semantic Segmentation Labels

A dataset of sidewalk images gathered in Belgium in the summer of 2021, intended for training semantic segmentation models. The dataset is labeled for the task of pixel-level classification, with model performance measured by mean intersection over union (IoU).

Use Cases

• Train a semantic segmentation model to classify each pixel in sidewalk images from Belgium.
• Benchmark model performance on the semantic-segmentation task using the mean IoU metric.
• Analyze sidewalk features and infrastructure patterns from images collected in a specific geographic region and time period.

Strengths

• Images were gathered in a specific geographic region (Belgium) and time period (summer 2021), providing contextual consistency.
• Dataset size is categorized as between 1,000 and 10,000 samples (1K<n<10K).
• Annotations were created through a combination of crowdsourced and expert-generated methods.

Limitations

• The dataset size is relatively small for deep learning, categorized as between 1,000 and 10,000 samples.
• Geographic coverage is limited to Belgium, which may limit model generalizability to other regions.
• Annotations may contain noise or inconsistencies due to the crowdsourced component of the labeling process.

Provenance

Source

segments.ai / Hugging Face user 'segments'

Collection Method

Images gathered in Belgium; annotations created via crowdsourced and expert-generated methods.

Time Range

Summer 2021

Freshness

Data was collected in summer 2021, with the dataset card last updated in July 2023.

Geography

Belgium