MEDFISH101: 70,000 Validated Images of 101 Mediterranean Fish Species

The Mediterranean Sea is the focus of this dataset of approximately 70,000 validated images covering 101 fish species, compiled to support automated species identification. Created by Andreas Kofidis and last updated in May 2026, the resource was used to develop deep learning pipelines achieving high accuracy for marine biodiversity monitoring.

Use Cases

• Train automated fish species classifiers based on the 101 labeled species.
• Benchmark vision foundation models for ecological image recognition based on the validated image collection.
• Support citizen science biodiversity monitoring in the Mediterranean Sea based on the dataset's geographic focus.
• Apply transfer learning techniques to marine biology tasks based on the described deep learning pipeline development.

Strengths

• Approximately 70,000 validated images provide a substantial volume for model training.
• Covers 101 distinct Mediterranean fish species, offering a broad taxonomic scope.
• The associated model achieved a Top-1 accuracy of 94.12%, indicating high utility for the intended task.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Data may reflect geographic bias inherent to its Mediterranean Sea focus.

Provenance

Source

figshare

Collection Method

Compiled and curated from citizen science initiatives, likely for the described study.

Freshness

Last updated 2026-05-20 10:52:40; freshness should be verified.

Geography

Mediterranean Sea