Questions for "Understanding the Limitations of CNN-based Absolute Camera Pose Regression"
-----------------------------------------------------------------------------------------------
Please send your answers to: schroepp@cs.uni-freiburg.de by 10:00 on 03.07.2019

1. The authors distinguish between 4 different approaches for visual localization: structure-based, image retrieval, absolute camera pose regression (APR) and relative camera pose regression (RPR). Give a short description of each of them and explain why it is problematic if learning-based approaches estimate poses similar to image retrieval rather than structure-based approaches. (4 sentences)

2. How is a predicted pose for an input image constructed from base poses in the authors' theoretical model of APR? Why is training data that was captured on a line potentially problematic in this case? (2-3 sentences)

3. What are the conclusions from the results of the different approaches on the Cambridge Landmarks and 7 Scenes datasets (Table 2)? (2-3 sentences)