Developer here. When I started studying deep learning and realized that the datasets that I wanted to learn on didn't exist. The more I looked at it, the more I realized what a constraint not having data was.
I started building an image cleaning tool, it evolved over time into a labeling tool. That got its first trial at an employer where I used it to run custom Mechanical Turk tasks. Then after experiencing the pain of mturk, iterated on it to be a stand alone tool. A lot of the features were my wish list from working with labeling contractors.
MekaBytes is now live as a labeling site. My vision is that people will be able to come together and contribute to datasets that they care about, similar to subreddits. Rather than them getting abandoned after the creator finishes their PhD or funding runs out, they'll get to live on and others can contribute and improve them over time.
Anyway, it's still early so feedback is very appreciated! Next up on my TODO list are comment notifications to make collaboration smoother.
1 comment
[ 1.9 ms ] story [ 16.5 ms ] threadI started building an image cleaning tool, it evolved over time into a labeling tool. That got its first trial at an employer where I used it to run custom Mechanical Turk tasks. Then after experiencing the pain of mturk, iterated on it to be a stand alone tool. A lot of the features were my wish list from working with labeling contractors.
MekaBytes is now live as a labeling site. My vision is that people will be able to come together and contribute to datasets that they care about, similar to subreddits. Rather than them getting abandoned after the creator finishes their PhD or funding runs out, they'll get to live on and others can contribute and improve them over time.
Anyway, it's still early so feedback is very appreciated! Next up on my TODO list are comment notifications to make collaboration smoother.