Today, I downloaded the model weights for a state-of-the-art R-CNN architecture, ran some object recognition on some pictures. Took me 15 minutes to install the

Today, I downloaded the model weights for a state-of-the-art R-CNN architecture, ran some object recognition on some pictures. Took me 15 minutes to install the dependencies (bunch of one-liners), modify the code a little, and get it running on my data; all the internals are explained in the papers, and can be modified in the scripts. Just saying, could be worse. To be fair, I agree that the big papers mostly come from google & MS, and I wish research was a real democracy.