Questions for the seminar Paper "Discovering States and Transformations in Image Collections"
-----------------------------------------------------------------------------------------------
Please send your answers to: bravoma@cs.uni-freiburg.de by 14:00 on 18.01.2022

1) What challenges does attribute detection poses compared to object or scene classification? Why is it important to work on attribute discovery? (2 sentences)

2) The method proposed achieves a lower performance on object categories as compared to materials or scenes. The authors argue this difference is due to the number of synonyms of the nouns. Do you think the correlation of these factors is shown in figure 9(b)? What other explanation can you think of for the difference in performances? (3 sentences)

3) This paper was proposed some years ago. With the knowledge and tools of the field, how would you approach the problem of zero-shot combination discovery? (unseen noun-attribute discovery) (3 sentences)