https://mshields.name/tags/machine-learning/ Recent content in machine learning on Matthew Shields Hugo -- gohugo.io © 2022 - 2026 Matthew Shields Fri, 19 Jan 2024 00:00:00 +0000 https://mshields.name/blog/2024-01-20-setting-up-cuda-11-8-and-pytorch-on-ubuntu-20-04-with-secure-boot-enabled/ Fri, 19 Jan 2024 00:00:00 +0000 https://mshields.name/blog/2024-01-20-setting-up-cuda-11-8-and-pytorch-on-ubuntu-20-04-with-secure-boot-enabled/ Preamble I recently got a fresh daily drive laptop which happens to have secure boot enabled. Typically this has added yet more complication to getting setup with CUDA and PyTorch. Hopefully the following reference can help someone else out or even just me if I need it again in the future. How To 1. Purge system of NVIDIA CUDA in case of a previous failed install sudo rm -r /var/lib/dkms/nvidia sudo rm /etc/apt/sources. https://mshields.name/blog/2022-03-10-registration-methods-comparison/ Thu, 24 Mar 2022 00:00:00 +0000 https://mshields.name/blog/2022-03-10-registration-methods-comparison/ Preamble Registration is the technique of aligning two data sets by finding either a rigid or affine transformation, depending on the problem. In robotics this is often between two pointclouds but the technique can be useful for aligning any data sets especially those that contain spatial data, MRI scan images for example. The Wikipedia page for this topic is quite concise and well written. Background Typically we represent transformations in 3D space as a matrix and translation vector that we apply to an input vector of coordinates to produce a transformed output vector of shifted coordinates. https://mshields.name/blog/2022-02-23-extracting-road-markings-from-pointcloud-data/ Sat, 26 Feb 2022 00:00:00 +0000 https://mshields.name/blog/2022-02-23-extracting-road-markings-from-pointcloud-data/ Preamble This is the first post in a series looking back at past projects I have done but not shared publicly or documented in any way. This was a piece of work from 2019 with the goal of extracting road markings from geo-referenced pointclouds. For those that don’t know, a geo-referenced pointcloud is created by taking a LiDAR and putting it on some kind of a rover vehicle, then taking all the observations of the LiDAR that were taken in the moving vehicle reference frame and converting them into the global “static” reference frame.