Dataset Overview

Dataset Name

The FungiTastic Dataset

Dataset Description

FungiTastic is a comprehensive multimodal machine‑learning dataset for classifying fungi from images and metadata. It includes photographs of fungal observations, satellite imagery, weather observations, segmentation masks, and textual metadata. Metadata enriches each observation with timestamps, camera settings, GPS location, substrate, habitat, and biological taxonomy. By combining multiple modalities, the dataset provides a robust benchmark for multimodal classification, enabling the development and evaluation of complex machine‑learning models under realistic and dynamic conditions.

Dataset Content

• Image Data: Photographs of fungal observations, satellite images, and segmentation masks.
• Metadata: Timestamps, camera configuration, GPS coordinates, substrate, habitat, and biological taxonomy.

Dataset Subsets

• FungiTastic Closed Set: Training (246,884), validation (45,616), test (48,379) observations.
• FungiTastic‑M Closed Set: Small prototype subset, training (25,786), validation (4,687), test (5,531).
• FungiTastic‑FS Closed Set: Few‑shot subset, training (4,293), validation (1,099), test (998).
• FungiTastic Open Set: Training (246,884), validation (47,453), test (50,085) observations.
• FungiTastic‑M Open Set: Small open‑set subset, training (25,786), validation (4,703), test (5,587).

Dataset Statistics

• Total Images: Over 650,000.
• Total Observations: Over 350,000.
• Class Distribution: Long‑tail distribution (see Figure 2 in the paper).

Evaluation and Metrics

The dataset defines five problem settings with corresponding evaluation metrics:

1. Fine‑grained closed‑set classification with a heavy long‑tail distribution.
2. Standard closed‑set classification with out‑of‑distribution (OOD) detection.
3. Classification with non‑standard cost functions.
4. Classification on a time‑sorted dataset for benchmarking adaptation methods.
5. Few‑shot classification for species with limited training observations.

Baseline Results

Performance metrics for various architectures on each subset are provided, including Top‑1, Top‑3, and macro‑averaged F1 scores.

Dataset Download

• Kaggle: FungiTastic
• GitHub: FungiTastic Repository