One thing I am very proud of: my very first dataset creation. I made a dataset (alongside with my project colleague) for our master thesis project. It contained more than 200.000 images of us watching TV.
We used the open source OpenPose (https://github.com/CMU-Perceptual-Computing-Lab/openpose) to extract the features that we later on used to feed our program (which was just a neural network) to classify three different levels of attention levels.
I was very embarrassed when extracting the frames after filming us but it was a cool experience and hey! if you are lacking data for any idea you have, is there a better way to obtain it than just creating it yourself? 馃檲
This is actually something that not a lot of people want to do when working in AI but it is very crucial. A big part of my day I spend it scarping data that would help solve the business problems.
