# Forams Classification 2025 — Data

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

采集时间：2026-10-18T13:41:19Z

# Foram Dataset and Submission Guidelines
The Forams Dataset contains 18,426 volumes of 128-by-128-by-128 voxels. This is a semi-supervised classification challenge, so only a few labeled samples are given. The data contains 210 labeled volumes and 18,216 unlabeled volumes. The task is to classify the unlabeled volumes into one of the labeled species or an unknown class. The unknown class should be used for broken fragments of forams and other unclassifiable objects, which may occur in unlabeled volumes. However, no labeled examples of the unknown class are provided.

# Folders:
**Volumes** - contains all the volumes split into two folders called 'labelled' and 'unlabelled'. Each volume file name contains the scaling factor used when scaling the files, with underscore used in the file name instead of the decimal point. 

Filenames for the unlabeled volumes:  
`foram_[5 digit id]_sc_[1 digit integers scale factor]_[3 digits decimal scale factor].tif` 

Filenames for the labeled volumes:
`labelled_foram_[5 digit id]_sc_[1 digit integer scale factor]_[3 digits decimal scale factor].tif` 

For example, a file name `foram_09707_sc_0_673.tif` contains:
 - id: 9707
 - scaling factor: 0.673. This means that the largest side of the original volume was 128/0.673 = 190 voxels. In the original volume, voxels are 3.7 micro-meter, so the physical side length of the scaled volume is approximately 0.7 mm (190 voxels times 3.7 micon/voxel).

**Visualizations** - contains one RGB image for each volume, showing a volume rendering and a surface rendering of the same volume. 'Visualizations' is structured similarly as 'Volumes', it contains two folders, 'labelled' and 'unlabelled'.

Filenames for images: 
`foram_[5 digit id].jpg` for the files in the 'unlabelled' folder and `labelled_foram_[5 digit id].jpg` in the 'labelled' folder.
For example, a file name `labelled_foram_00017.jpg` contains:
 - file id: 17
 - rendering of the `labelled_foram_00017_sc_0_680.tif` file

**labelled.csv** - contains two columns. The first is the id of the files, and the second is the label of each of the 210 volumes included in the labelled set. The order is the same as in the 'labelled' folders in the `Volumes` and `Visualization` folders. The labels 0 to 13 come in sorted order, such that the first 15 rows are for files with label 0, the next 15 are for label 1, etc.

**unlabelled.csv** - contains two columns. The first is the id of the files, and the second is empty and should be filled out for submission. The labels must be 0 for the first class, 1 for the second, etc. The label for unknown must be 14.

**sample_submission.csv** - example of a submission.

# Evaluation
Predictions are evaluated by matching the predicted label to the true labels and calculating the F1-score. The evaluation is based on the `f1_score` from scikit-learn and is used to calculate the F1 score between the ground truth labels and the submission.