Sergey Mushinskiy tells his solution of Kaggle Carvana Image Masking Challenge. In this competition, Carvana is challenging Kagglers to develop an algorithm that automatically removes the photo studio background. From this video you learn:
Using pseudo-labeling to train network with predicted test samples
Teamwork with shared GPUs time
Is it useful to hand label
Other approaches: from two network without any averaging to complex ensemble with insane architectures