# AI-KFM 2022 — Data

来源：https://www.kaggle.com/competitions/ai-kfm-2022/data

本地 Meta Kaggle 快照；不是本次实时页面。

# Dataset Description

The challenge uses image data samples saved in .png files. The dataset consists of 11 different acquisitions: seven for the Train set and 4 for the Final test set. 
Each acquisition in the Train set has two folders:

*   **images** : containing the images (.png files) belonging to the same acquisition. Each filename has a specific structure, that is micro_xx_yy.png, where xx and yy are the information used to compose a single image for each acquisition. It means that the images can be considered as patches (partially overlapping) extracted from a bigger image;
*   **masks**: containing the coordinates of the boxes surrounding the eggs, saved as .roi files. This means that, if the image micro_xx_yy.png in the images folder contains an egg, a micro_xx_yy.roi file is placed in the masks folder. If the micro_xx_yy.png file contains more than one egg, the corresponding micro_xx_yy.zip includes a .roi file for each egg. For each .roi file, the field `vnRectBounds` reports the four coordinates of the box as  `X_min, Y_min, X_max, Y_max`.

The acquisition in the Test set contains only the images folder. 
## Files

*   **Train.zip** - the training set
*   **Final.zip** - the final test set, consisting of the previously released test set (1 scan) plus new images (3 scans). Please note that new and old images can have different sizes
*   **SampleSubmission_final.csv** - a sample submission file in the correct format, for the final test set
*   **Submission Example Generation** - a zip file containing a full working MATLAB example to create a submission file from per-image binary masks