Questions for
"Hollywood in Homes: Crowdsourcing Data Collection for Activity Understanding" by Gunnar A. Sigurdsson et al.

Send your answers to zolfagha@cs.uni-freiburg.de before July 20th, 2016, 10am.

1) What is the motivation and advantage of creating this dataset? What are problems of other datasets (UCF101, ActivityNet, Sport1M,..)? (Max 4 sentences)
2) How do the authors ensure diversity and generalization of collected videos? (~2-3 sentences)
3) Do you think that actions with higher number of examples tend to have higher accuracy? What do the authors conclude based on the confusion matrix presented in Fig. 7? (~1-3 sentences)