data

Data

Datasets

Training

• Lab
• Auxiliary

Test set for image registration

• FIRE: https://projects.ics.forth.gr/cvrl/fire/

Test sets for identity verification

• VARIA: http://www.varpa.es/research/biometrics.html
• BES
• CLINICAL

Data Organization

Training data

Our semi-supervised training code reads a labeled training dataset (Lab in our ECCV paper) and an unlabeled training dataset (Auxiliary), and assumes each training data to be organized as follows.

Lab/
    Annotations/
        vistel0_left_0.txt
        vistel0_left_1.txt
        ....       
    ImageData/
        vistel0_left_0.jpg
        vistel0_left_1.jpg
        ....     
    ImageSets/
        eccv22_train.txt  
        eccv22_val.txt    
        Lab.txt  
Auxiliary/
    image1.jpg
    image2.jpg
    ...

• The Annotations folder contains keypoint annotations per image. The folder is optional for unlabeled data. A sample annotation file is given as samples/vistel0_left_0.txt. See the tutorial code that explains how the keypoint annotations shall be stored and loaded.
• The ImageData folder contains all image files.
• The ImageSets folder contains image-id files that specify data split. See eccv22_train.txt, eccv22_val.txt and lab.txt

Test data

The file organizations are as follows:

FIRE/
    Ground Truth/
    Images/
    Masks/

VARIA/
    Images/
        R001.pgm
        R002.pgm
        ...
    pair_index.txt

• Note that the annotation file of control_points_P37_1_2.txt in FIRE dataset is incorrect, so it shall be excluded from evaluation.

For identity task, pair_index.txt is used to indicate matching pairs of the dataset.

# each line in the index file has three colunms, means
query_image, refer_image, 0 (reject) or 1 (accept)
# for example
R180.pgm, R002.pgm, 1
R012.pgm, R002.pgm, 0