Writer Verification Challenge 2023

Challenge

Welcome to the Summer Challenge on Writer Verification, hosted as part of the National Conference on Computer Vision, Pattern Recognition, Image Processing and Graphics (NCVPRIPG'23). Given a pair of handwritten Hindi Text images, the task is to automatically determine whether they are written by the same writer or different writers.

Writer Verification task originated as a signature verification problem to flag potentials frauds in banking-sector. This is a challenging task as the individual handwriting can significantly vary from person to person and thus the underlying model needs to learn the varations in the handwriting style. The problem becomes even more complex in the offline setting, where the dynamic information about the writing process is not available (like writing in an electronic equipment).

Evaluation metrics

Following the literature on verification tasks, we shall use the following performance measures:

• F1
• AUC

The F1 score is a measure of a model's accuracy that takes into account both precision and recall.

The Area Under the Curve (AUC) is a metric that measures the performance of a binary classification model by calculating the area under the receiver operating characteristic (ROC) curve. The AUC ranges from 0 to 1, where a score of 1 indicates a perfect classifier, while a score of 0.5 indicates a random classifier. Note that the Leaderboard will be decided based on the AUC score only.

Input and output specifications

Training set contains 1352 folders of images. All the images in one folder are written by the same person. Validation set contains a set of images from 92 different writers. The val.csv file contains name of image pairs and the corresponding labels. A label of 1 indicates that the images are written by the same writer and a label of 0 indicates that the images are written by different writers.
For futher details, please refer to the Github repository.

Test set contains images from 360 writers. In test.csv you are given name of image pairs. For output, you need to predict the label for given pair of images and submit the csv file in the format by editing the test.csv file. Note that test set will be released at a later date than train and validation set.

Code to get started

A github repository is provided to get started with the challenge. The repository contains a sample code to read the dataset and evaluate the performance and other necessary instructions.

Events schedule

When: July 21, 2023; 1730-1930 IST