Aerial Elephant Detection Benchmark with 15,511 Annotations

A collection of 2,101 aerial images containing 15,511 annotated elephants, split into training and test sets. The dataset was created by J. L. P. Naudé from images acquired over 8 separate campaigns in different environments. Image resolution varies between 2.4 cm/pixel and 13 cm/pixel, with nominal resolution specified in metadata.

Use Cases

• Benchmarking object detection algorithms based on the provided training and test split.
• Developing models for wildlife population monitoring based on aerial imagery.
• Studying the impact of varying ground sample distance (GSD) on detection performance based on the provided resolution metadata.
• Training models for conservation applications based on the focus on elephants.

Strengths

• Contains 15,511 annotated elephant instances across 2,101 images, providing a substantial number of examples.
• Explicit train/test split (1,649/452 images) designed to maintain independence by not splitting clusters of related images.
• Image resolution metadata is provided, allowing for consistent resampling.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Last update date is unknown; freshness unverified.
• Specific geographic locations of the 8 campaigns are not detailed, which may limit understanding of environmental diversity.

Provenance

Source

J. L. P. Naudé via paperswithcode.

Collection Method

Images were acquired over 8 separate aerial campaigns.