We are providing still frame images extracted from webinar videos containing presentation slides with one or more titles. The objective of the challenge is to identify the title(s) in each slide that would be used to annotate the presentations.

Be aware that some training slide images may contain up to 4 titles at once, which is reflected in Train.csv ("Title1", "Title2", .., "Title4" columns). This is done to enrich your training with as much title samples as possible. Note that all our test images can only have a single title and you are expected to predict it.

The Train.csv and SampleSubmission.csv comprise the ID and Title columns where:

• The labels in the ID columns correspond to the image labels without the leading zeros, i.e. image 000537.jpg in the train set corresponds to ID 537 in the Train.csv file.
• The Title variable(s) represents the ground truth, i.e. title, of each slide/image in both the train and test set