datasets
datasets copied to clipboard
Feature: BOBSL dataset integration
Complete work on https://github.com/sign-language-processing/datasets/pull/5
https://www.robots.ox.ac.uk/~vgg/data/bobsl/ is the dataset website
I will probably take on this as I am working more on BOBSL.
The first step for me is to construct a BOBSL_ISLR dataset consisting of ~5M huge but noisy sign-level training examples.
I will then look at the subtitle/sentence-level based on #5 if I work on MT. Apparently, VGG has their own pipeline, a bit messy, will see how to integrate everything efficiently.