The training and test images for the competition were all collected in the Monterey Bay Area between the surface and 1300 meters depth. The images contain bounding box annotations of 290 categories of bottom dwelling animals. The training and evaluation data are split across an 800 meter depth threshold: all training data is collected from 0-800 meters, evaluation data comes from the whole 0-1300 meter range. Since an organisms' habitat range is partially a function of depth, the species distributions in the two regions are overlapping but not identical. Test images are drawn from the same region but may come from above or below the depth horizon. 

The competition goal is to label the animals present in a given image (i.e. multi-label classification) and determine whether the image is out-of-sample.    

## Data format
The training dataset are provide in two different formats: multi-label classification and object detection. The different formats live in the corresponding named directories. The datasets in these directories are identical aside from how they are organized. 

### Multi-label classification
Each line of the csv files indicates an image by its `id` and a list of corresponding `categories` present in the frame. 
```
id, categories
4a7f2199-772d-486d-b8e2-b651246316b5, [1.0]
3bddedf6-4ff8-4e81-876a-564d2b03b364, "[1.0, 9.0, 11.0, 88.0]"
3f735021-f5de-4168-b139-74bf2859d12a, "[1.0, 37.0, 51.0, 119.0]"
130e185f-09c5-490c-8d08-641c4cbf6e54, "[1.0, 51.0, 119.0]"
```
The `ids` correspond those in the object detection files. The `categories` are the set of all unique annotations found in the associated image.

### Object detection
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_id` ranging from 1 to 290 and a `supercategory` from 1 to 20. The fine-grained annotations are taxonomic thought not always at the same level of the taxonomic tree. 

### Supercategories
Each category also belongs to one of 20 semantic supercategories as indicated in `category_key.csv`:
```
['Worm', 'Feather star', 'Sea cucumber', 'Squat lobster', 'Fish', 'Soft coral', 'Urchin', 'Sea star', 'Sea fan', 'Sea pen', 'Barnacle', 'Eel', 'Glass sponge', 'Shrimp', 'Black coral', 'Anemone', 'Sea spider', 'Gastropod', 'Crab', 'Stony coral']
```
These supercategories might be useful for certain training procedures. The supercategories *are* represented in both the training and validation set. But please note that submissions must be made identifying the 290 fine grained categories. 

We are not able to provide images as a single downloadable archive due to [FathomNet's Terms of Use](https://fathomnet.org/fathomnet/#/license). Images should be downloaded using the indicated `coco_url` field in each of the annotation files. Participants can either write their own script or use the provided `download_images.py` python script. 

## 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 full [FathomNet Use Policy](https://fathomnet.org/fathomnet/#/license). 

## Files

*   **multilabel_classification/train.csv** - csv list of training images and categories
*   **object_detection/train.json** - the training images, annotations, and categories in COCO formatted json
*   **object_detection/eval.json** - the evaluation images in COCO formatted json
*   **sample_submission.csv** - a sample submission file in the correct format
*   **category_key.csv** - key mapping numerical index to category and supercategory name
* **demo_download.ipynb** - python notebook demonstrating download script
*  **download_images.py** - python script to download imagery from FathomNet using COCO formatted json 
* **requirements.txt** - python requirements to run the download script

## 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, create a virtual environment running `python 3.9`:

```
$ conda create -n fgvc_test python=3.9 pip
$ conda activate fgvc_test
$ pip install -r requirements.txt
```

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

```
$ python download_images.py [PATH/TO/DATASET.json] --output [PATH/TO/IMAGE/DIRECTORY]
```

If no `--output` directory is specified, the script by default downloads images to the directory the command is executed from.