BING: Binarized Normed Gradients for Objectness
===============================================
- The authors categorize the related research into three categories. What are they? What are some of the shortcomings of the approaches in the two other categories while compared to Objectness proposal generation? What do you think is the next best proposal generator after BING and why?
- List some of the advantages and disadvantages of using NG features. How do the authors propose to overcome the disadvantages in the final detection-classification pipeline?
- How do the authors improve the speed of the NG feature extraction process? What is the main challenge of their proposed speed up approach? Will BING provide reasonable bounding boxes for every conceivable object category, explain?

Seeing 3D chairs: exemplar part-based
===============================================
- What are the limitations of 3D model-based methods and of 2D appearence-based methods? How do the author overcome those limitations?
- Highlight the main contributions of the work and briefly describe the proposed algorithm.
- What is meant by calibration of the score and why is it critical?

PANDA: Pose Aligned Networks
===============================================
- What are convolutional neural networks?
- What is Attribute classification?
- What are pose aligned networks?

Tracking Interacting Objects Optimally
===============================================
- What is the main contribution of this paper?
- What is mixed integer programming?
- Why are flow constraints needed and important?

SLAM++ Simultaneous Localisation and Mapping
===============================================
- What is the main novelty of the paper and how it differs from the previous works on SLAM?
- What are the assumptions of the work and how can they be violated? What did the authors do to make the approach robust to assumption violations?
- Which kind of prioir knowledge can the algorithm use and how is it exploited?

All-Environment Visual Place Recognition with SMART
===============================================
- The authors based their algorithm on their previous work called SeqSLAM. Which are the main new aspects and how these improve the performance of the previous work?
- Briefly describe the SMART algorithm pipeline. Why do they use a non-linear image sequence searching?
- What are the main issues they address in the two datasets they used? How the sequence lenght affects the performance in the road dataset and how do they address issues from changes in vehicle speed?

Mobile Robot Navigation System in Outdoor Pedestrian Environment
===============================================
- The authors compare their road recognition system with 3 other benchmarks. Breifly describe these 3 benchmark approaches in a sentence or two, and what are the disadvantages of each these approaches as identified by the authors of this paper?
- Briefly describe the two ways to recognize the road after the edge detection step as mentioned in this paper? Which pipeline is better and why?
- What is the purpose of introducing a horizon support line in the road line extraction algorithm? Can road lines be detected without this?

Pulling Things out of Perspective
===============================================
- Describe the main limitations of the state-of-the-arts methods the authors summarize for the semantic segmentation domain. Which aspects are they addressing in this work?
- Describe the property of "perspective geometry" and how they want to use it to create an unbiased depth classifier for the depth estimation problem.
- Why and how do the authors condition the semantic labeling with the depth estimation problem? What are the main advantages/weaknesses of their new joint pixel-wise classifier?