Bangla-Captioning-Image-Taken-By-Blind-People
Blind people are an important part of our society. Due to technological revelation, it is easy enough to use smartphones and captures images by voice or semi/fully guided command. It will be helpful to caption the captured image and describe the image to blind people. Our target is solving this problem with computer vision, machine learning, and image processing.
The idea originates from Gurari et al who are responsible for creating a dataset with 39,000 images originating from people who are blind that are each paired with five English captions. We are going to build a similar dataset for the Bangla language. For creating the dataset, we are considering two ways. One is using crowd workers and another is using translator API. At first, we are going to use translator API to build the pipeline for predicting captions from the images captured by blind people. In parallel, our crowd workers will caption the image in the Bangla language. After building the dataset, we are likely to build our pipeline with pre-trained models for captioning the image.
Currently, we are building our dataset pipeline. We are hoping to complete and generating a result within November, 2020.