# FungiCLEF25 @ CVPR-FGVC & LifeCLEF — Data

来源：https://www.kaggle.com/competitions/fungi-clef-2025/data

本地 Meta Kaggle 快照；不是本次实时页面。

# Dataset Description

# About
The training and validation dataset is built from fungi observations (i.e., multiple photographs of the same specimen and additional observation metadata) submitted to the Atlas of Danish Fungi before the end of 2023, which were labeled by mycologists. Beyond photographs, the dataset includes a wealth of supplementary data, such as satellite imagery, meteorological records, and structured metadata (e.g., timestamps, GPS coordinates, substrate details, habitat information, and toxicity status). The vast majority of observations have been annotated with most of these attributes.

## Structure
* **images/** - folder with the dataset images, available in different sizes: 300px, 500px, 700px and original.
* **metadata/** - folder with separate metadata files for the training, validation and test dataset splits.
* **captions/** - folder with separate caption for each image saved as .json
* **FungiCLEF25-SAMPLE_SUBMISSION.csv** - a sample submission file in the correct format

## Columns

*   `ObservationID` - definition of example_id
*   `categoryID` - definition of feature_1
*   etc., see Tabular data description

# Statistical overview of dataset splits
All of the images are also accompanied by tabular metadata and automatically-generated text descriptions of the images. Each of the classes has between 1-4 observations in the training set.

| Split            | Observations      | Images        | Classes      | 
|--------------------|-------------|-------------|
| Training             | 4,293       | 7,819       | 2,427     | 
| Validation               | 1,099       | 2,285       | 570         | 

# Captions

For each photograph, we provide a Malmo-7B image caption with an exhaustive text description.

# Tabular data 

| Metadata               | Description                                                                                                     |
|------------------------|---------------------------------------------------------------------------------------------------------------|
| **Observation date**   | Date when the specimen was observed in yyyy-mm-dd format. Besides, three additional columns with pre-extracted year, month, and day values are provided. |
| **EXIF**               | Camera attributes extracted from the image, e.g., metering mode, color space, device type, exposure, etc.      |
| **Habitat**            | The environment where the specimen was observed. Selected from 32 values such as *Mixed woodland*, *Deciduous woodland*, etc. |
| **Substrate**          | The natural substance on which the specimen lives. Selected from 32 values such as *Bark*, *Soil*, *Stone*, etc. |
| **Taxonomic labels**   | For each observation, we provide full taxonomic labels that include all ranks from species level up to kingdom. All are available in separate columns. |
| **Toxicity status**    | Whether the species is toxic or not. Since non-edible species can cause serious health issues as well, we label them as poisonous. |
| **Location**           | Location GPS data, also converted to various formats. Besides the latitude and longitude, we provide coarser administrative divisions into regions, districts, and countries. |
| **Biogeographical zone** | One of the major biogeographical zones, e.g., *Atlantic*, *Continental*, *Alpine*, *Mediterranean*, and *Boreal*. |
| **Elevation**          | Standardized height above the sea level.                                                                      |
| **Landcover**          | Land cover classification code with values like *savanna*, *barren*, etc. Taken from MODIS Terra+Aqua.   |