Questions for "COBE: Contextualized Object Embeddings from Narrated Instructional Video" ----------------------------------------------------------------------------------------------- Please send your answers to: bravoma@cs.uni-freiburg.de by 10:00 on 07.01.2020 1. What is the procedure the authors used to build the annotated data set? What are the drawbacks and challenges of this method? (2-3 sentences, numerate steps) 2.What do the authors mean when they say that they want to produce "contextualized object embeddings" and how do they force this context? (2-3 sentences) 3. Why do they use contrastive loss instead of traditional cross-entropy loss to learn object embeddings? What is the benefit of using contrastive loss? (2-3 sentences)