Register Here

Objective:  

To create an image annotation portal, to annotate images of the handwritten text of children to be used in custom training a Hand Written Text Recognition model for Indian Regional Languages

 

Concept:

Volunteers would need to create a web application portal in which images of the handwritten text of children could be uploaded and volunteers would be able to annotate those images on the word level.

 

Issues Addressed:

Assessments and Evaluations play an important role in the student-teacher exchange process as well as the overall learning process. Conventionally, answer evaluations involve a lot of manual processes on the part of the teacher. A teacher has to prepare the assessment for a topic, collect student answer sheets and evaluate them manually to give feedback to the students. The automation of this assessment process would reduce this manual work, and help teachers focus more on course content delivery and preparation of good assessments for students. Assessments form an integral part of the work being done at Pratham. We have developed assessment apps in which students can submit their handwritten text answer images for an assessment. The first step towards automating the assessment process would be to extract text from these images. Most of the images are in Indian Regional languages. To create an OCR tool for handwritten text recognition for Indian regional languages we need to annotate existing data for training a Hand Written Text Recognition model.  This annotation portal will help us create a dataset to train our models.

Activity Details:

  • Building a web application portal for annotating handwritten text in Indian Regional Languages
  • The portal should preferably automatically create bounding boxes around words in the images that are uploaded to the portal using available models (CRAFT, EAST, etc). These bounding boxes should be editable, for volunteers to adjust and annotate
  • There should be options to rotate the images since most images are submitted in different camera angles
  • The Annotated data should be downloadable in CSV, Excel, and JSON format
  • The web application should also contain a dashboard that gives details regarding metadata such as number of images annotated, number of images annotated by language, number of images annotated by users, and so on
  • Once the web application is hosted, we can upload the data and get started with the annotation work.
  • Reference Tool- VGG Image Annotator (ox.ac.uk)
Register Here
Hand Written Text Image Annotation Portal

Please write to us at digital@pratham.org to join us in our efforts to create equitable learning for children across the country.