Motivation of 3D Scene Understanding

Some extract from papers, but may be minorly modified by me.

Data-Driven 3D Primitives for Single Image Understanding

How do you infer the 3D properties of the world from a 2D image? This question has intrigued researchers in psychology and computer vision for decades. Over the years, researchers have proposed many theories to explain how the brain can recover rich information about the 3D world from a single 2D projection. While there is agreement on many of the cues and constraints involved (e.g., texture gradient and planarity), recovering the 3D structure of the world from a single image is still an enormously difficult and unsolved problem.
    At the heart of the 3D inference problem is the question: What are the right primitives (representations) for inferring the 3D world from a 2D image? It is not clear what kind of 3D primitives can be directly detected in images and be used for subsequent 3D reasoning. There is a rich literature proposing a myriad of 3D primitives ranging from edges and surfaces to volumetric primitives such as generalized cylinders, geons and cuboids. While these 3D primitives make sense intuitively, they are often hard to detect because they are not discriminative in appearance. On the other hand, primitives based on appearance might be easy to detect but can be geometrically uninformative.
    They propose geometric primitives which are visually-discriminative, or easily recognized in a scene, and geometrically-informative, or conveying information about the 3D world when recognized.

GeoNet: Geometric Neural Network for Joint Depth and Surface Normal Estimation

Albeit the great advancement in this filed (depth estimation), we notice that most previous methods deal with depth and normal estimation independently, which possibly make their prediction inconsistent without considering the close underlying geometry relationship. For example, as demonstrated in [], the predicted depth map cloud be distorted in planar regions. It is thus intriguing to ask what if one considers the fact that surface normal does not change much in planar regions. This thought motivates us to design new models, which are exactly based on above simple fact and yet potentially show a vital direction in this field, to exploit the inevitable geometric relationship between depth and surface normal for more accurate estimation.

undefined reference to `xcb_dri3*'

Problem:
/usr/lib/x86_64-linux-gnu/libGL.so: undefined reference to `xcb_dri3_buffers_from_pixmap

Analysis:
1. libGL.so not working
2. xcb not working

Solution:
1. check if libGL.so is working
ldd  /usr/lib/x86_64-linux-gnu/libGL.so
if working: go to step 2
if not: install opengl

2. add -lxcb -lxcb-dri3
Note that we need to add all libs in the target makefile or qt pro file. Actually, I fix this with this solution.

3. sudo apt-get update
add -lxcb -lxcb-dri3