Computer Vision News - December 2019

We Tried for You 42 A few weeks ago , we have seen together how to generate a 3-dimensional model from a collection of 2D images using Visual SfM. Although the final 3D model we generated was quite appealing, it was far from being an industry-level one. When dealing with a point cloud generated from sensors, there are several challenges such as outlier removal, registration, retrieval andmore. Since 3D objects might be very complex to handle, a special tool is needed. by Amnon Geifman To this end we tried for you the Open3D library in python. Open3D is an open source library that supports rapid development of software that deals with 3D data. The backend of the library is highly optimized and allows efficient usage of many 3D algorithms. Among the core feature of Open3D you can find: basic 3D data structures and algorithms, scene reconstruction, surface alignment, visualizations and more. In this article, we will familiarize you with a few useful features of this library. Open3D settings and I/O Open3D can be used in either C++ or python. If you are using the C++ version, you can check out which compilers are supported and compile it on your machine. Luckily enough, on python we can get Open3D using pip install so we just write in the command line: pip install open3d Most of the algorithms in Open3D are self-contained, so that we don't need to install any further libraries and we are ready to go. In this article we build upon the work we did in the VisualSfM article . There, we end up with 3D reconstruction, meaning we had camera locations and 3D points. We also explained how to parse the 3D model so you only need to copy the parser of the 3D points and save the result into a .ply file. You can also use as input any point cloud file you desire from any other Structure from Motion software. Open3D for 3D processing