Pl@ntNet-300K-v2: Plant Images for Set-Valented Classification

Pl@ntNet-300K-v2 comprises 306,087 images of 1,000 plant species derived from the Pl@ntNet citizen observatory. It was created by researchers including Camille Garcin from the Centre National de la Recherche Scientifique and described in a 2021 NeurIPS publication. The dataset is designed to benchmark set-valued classification methods due to its strong class imbalance and visual similarity between species.

Use Cases

• Benchmarking set-valued classification methods based on the dataset's stated design purpose.
• Training models for fine-grained plant identification based on the 1,000 species classes.
• Researching methods to handle class imbalance based on the described strong imbalance where a few species account for most images.
• Studying visual similarity and ambiguity in image classification based on the challenge of visually indistinguishable species.

Strengths

• Contains 306,087 images, providing a substantial volume of data.
• Covers 1,000 distinct plant species, offering a wide taxonomic scope.
• Includes metadata files with species taxonomic details and IUCN conservation status.
• Provides improved image resolution and species naming compared to a prior version.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Last update date is unknown; freshness unverified.
• Data may reflect geographic or temporal bias inherent to the Pl@ntNet citizen observatory source.

Provenance

Source

Pl@ntNet citizen observatory database.

Collection Method

Derived from citizen science observations.