# FathomNet 2025 @ CVPR-FGVC — Data

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

本地 Meta Kaggle 快照；不是本次实时页面。

# Dataset Description

## Data Format
The datasets are formatted to adhere to the [COCO Object Detection](https://cocodataset.org/#format-data) standard. Every training image contains at least one annotation corresponding to a category from 1 to 79.

We are not able to provide images as a single downloadable archive due to FathomNet Database's Terms of Use. Images should be downloaded using the indicated coco_url field in each of the annotation files. Participants can either use the provided download.py python script or write their own. The download script will save two folders: an image and a region of interest (ROI) folder. The ROI folder contains cropped images that are saved with the following file format <image_id>_<annoation_id>.png while the image folder contains files saved as <image_id>.png, where the ids are specified in the dataset.json file provided to download.py. 

## Files
- dataset_train.json - the training images, annotations, and categories in COCO formatted json
- dataset_test.json - the evaluation images in COCO formatted json
- sample_submission.csv - a sample submission file in the correct format
- download.py - python script to download imagery from FathomNet

## Terms of Use
By downloading and using this dataset you are agreeing to FathomNet's data use policy. In particular:
The annotations are licensed under a Creative Commons Attribution-No Derivatives 4.0 International License.
The images are licensed under a Creative Commons Attribution-Non Commercial-No Derivatives 4.0 International License. Notwithstanding the foregoing, all of the images may be used for training and development of machine learning algorithms for commercial, academic, and government purposes.
Images and annotations are provided by the copyright holders and contributors "as is" and any express or implied warranties, including, but not limited to, the implied warranties of merchantability and fitness for a particular purpose are disclaimed.
Please acknowledge FathomNet and MBARI when using images for publication or communication purposes regarding this competition. For all other uses of the images, users should contact the original copyright holder.
For more details please see the FathomNet Database [Terms of Use](https://database.fathomnet.org/fathomnet/#/license).

## Download

Images are made available for download by the unique URLs in the COCO-formatted object detection annotation files. The download script can be run from the command line. To install the requirements, run the following with Python >= 3.8:

```bash
pip install -r requirements.txt
```

To download the images, run the script from the command line:

```bash
python download.py path/to/dataset.json path/to/output_dir [-n NUM_DOWNLOADS]
```

The optional `-n`/`--num-downloads` parameter can be used to specify the maximum number of concurrent downloads. Setting this to a value >1 can speed up the total time it takes to download the images.

>💡 Tip
> If you've cloned this repository, the COCO dataset file is available at `dataset/dataset.json`:
> ```bash
> python download.py dataset/dataset.json dataset/
> ```