Dhaka Building Footprints with 30cm Satellite Imagery and 189,057 Labels

A 2023 dataset contains 11,905 high-resolution satellite image tiles and corresponding building footprint vector labels for Dhaka, Bangladesh. It was created by MLHUB and sourced from Maxar ODP imagery with a 30 cm resolution. This dataset is a reviewed 'Tier 1' dataset used to develop a fine-tuned model for building detection.

Use Cases

• Training building segmentation models based on high-resolution 30 cm satellite imagery.
• Analyzing urban density and land use patterns based on the 'Very Dense Urban, Rural, Agricultural, Forested' keywords.
• Fine-tuning pre-trained geospatial models for specific regional applications, as it was used to develop the 'ramp fine-tune Dhaka Model'.
• Validating automated building extraction algorithms against a large set of 189,057 manually reviewed footprint labels.

Strengths

• Contains 11,905 image/label tile pairs, providing substantial training data.
• Includes 189,057 individually labeled building footprints for model training and validation.
• Imagery has a high 30 cm resolution, sourced from Maxar ODP.
• Dataset is designated as 'Tier 1', indicating it has been thoroughly reviewed and improved.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Last updated 2023-01-01 00:00:00; freshness should be verified.

Provenance

Source

MLHUB via NASA Earthdata.

Collection Method

Satellite imagery sourced from Maxar ODP; building footprint labels were created and reviewed.

Freshness

Last updated 2023-01-01 00:00:00.

Geography

Dhaka, Bangladesh.