Downsized Camera Trap Images for Automated Species Classification

Borneo camera trap images resized to 256x256 pixels, used to analyze CNN generalizability across habitat variation. The dataset contains 270,287 image records, with images tagged as 'human' removed. It was collected as part of the SAFE research project funded by NERC and is released by Imperial College London.

Use Cases

• Training convolutional neural networks for automated species classification based on camera trap images.
• Evaluating model generalization across different habitat disturbance levels based on the 'dist_level' field.
• Benchmarking classification performance using the defined training, validation, and test splits mentioned in the description.

Strengths

• Dataset contains 270,287 image records.
• Images are pre-processed to a uniform 256x256 pixel size.
• Includes metadata on camera trap site, taxon, and disturbance level for each image.
• Explicitly defines multiple dataset splits for training, validation, and testing analyses.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is known, but the total number of unique image files is not specified.

Provenance

Source

SAFE research project, Imperial College London

Collection Method

Collected via camera traps as part of the project 'Machine learning and image recognition to monitor spatio-temporal changes in the behaviour and dynamics of species interactions'.

Geography

Borneo (inferred from 'BorneoCam' dataset name)