Category: Software

Orchestrating ML Success on Your Own Terms: The Benefits of RoadMentor’s Customer Infrastructure

The RoadMentor platform is an innovative machine learning operations (MLOps) platform specifically designed for advanced driver assistance systems (ADAS). This platform leverages the latest in infrastructure orchestration technologies to provide customers with a powerful and flexible solution for managing their ML models and workflows. In this blog, we will delve into the key features of…
Read more

The Benefits of a Unified Product Experience

The market for machine learning algorithms for autonomous driving is still relatively young and, as a result, remains highly fragmented. With so many companies and organizations developing their own unique solutions and technologies, one must go to one company for data collection, another for management, and others for model development, MLOps, and even deployment. It…
Read more

How to stitch multiple cameras together on a moving vehicle

Image stitching is the process of combining multiple images together to create a seamless panorama or a large-scale image. This technique is commonly used in photography, virtual reality, and mapping applications. In this blog, we will take a deep dive into the concepts of image stitching, including sensor and vehicle frame of reference, origin point,…
Read more

Revolutionizing L3 ADAS: Harnessing the Power of Test Harnesses and MLOps for Safe and Reliable Automotive Technology

Regression testing is a crucial aspect of developing L3 Advanced Driver Assistance Systems (ADAS) technology. It involves re-testing the entire system or parts of it, after changes or modifications have been made, to ensure that the changes did not introduce any new bugs or defects. This is particularly important for ADAS systems, which rely on…
Read more

What is ADAS (Level 2 and below)?

Advanced Driver Assistance System (ADAS) is a technology that helps drivers to drive safely and efficiently. It is a combination of sensors, cameras, and other technologies that work together to help drivers monitor their speed, maintain their lane, and avoid collisions. ADAS can be found in many modern cars, and it is becoming increasingly popular…
Read more

Revolutionizing the Road: How Hyperspec AI’s Fleet Learning Approach is Changing the Game for Detecting Lane Merges on the Highway

Automating the process of data annotation has been a long-standing challenge in the field of machine learning, especially when it comes to detecting lane merges on the highway. The traditional approach of manually labeling data is not only costly in terms of time and resources, but it also hinders the ability to quickly detect lane…
Read more

Uncovering ADAS Failure Cases: The Importance of Edge Vision Analytics and Fleet Queries in Vehicle Safety

Collecting failure cases from a fleet of L3 ADAS vehicles is a crucial task in ensuring the safety and reliability of these systems. With the increasing use of advanced driver assistance systems (ADAS) in vehicles, it is important to have a reliable and efficient method for collecting and analyzing data from these systems. One way…
Read more

Jane, a superstar ADAS engineer

As the sun rose over the bustling city, Jane, an Adas engineer at a leading autonomous vehicle company, woke up to the familiar sound of her alarm clock. She knew that today would be a challenging day, filled with long hours and intense problem-solving. Jane arrived at the office and immediately got to work on…
Read more

Javascript, Caching to Optimize Network Traffic

Browsers use caching to temporarily store resources locally on the device, such as HTML, CSS, and JavaScript files, images, and videos. This allows the browser to quickly access the resources without having to make additional requests to the server. One way browsers can enable device-side caching is through the use of local storage. Local storage…
Read more

Training PackNet-SFM with RGB + PointCloud Data

Fusing time synchronized RGB camera images with point cloud data is a powerful approach for training a PackNet-SFM (Structure from Motion) model in PyTorch. This technique combines the benefits of both RGB images and point cloud data to improve the accuracy and robustness of the SFM model. In this article, we will discuss the various…
Read more

Integrating Federated Learning into CVAT & MLFlow

Federated learning is a machine learning technique that enables the training of models on decentralized data, without the need for the data to be centralized in one location. Instead, data is distributed across a number of different devices or edge devices, such as smartphones or IoT devices, and the model is trained by aggregating updates…
Read more

WebGPU and Federated Learning with FedML, a Killer Combo

WebGPU is a new technology that allows developers to take advantage of the power of the GPU (graphics processing unit) in modern browsers. It allows for faster and more efficient processing of complex tasks, including machine learning algorithms. One of the key benefits of WebGPU is its ability to support federated learning. Federated learning is…
Read more

Training a Model in MLFlow from CVAT label data

CVAT (Computer Vision Annotation Tool) is an open source tool developed by Intel that allows users to label and annotate images and video data for training machine learning models. MLFlow is an open source platform for managing the end-to-end machine learning lifecycle. It provides tools for tracking experiment runs, organizing code, and reproducing runs, among…
Read more

Integrating CVAT annotation into MLFlow

CVAT (Computer Vision Annotation Tool) is an open-source annotation tool for computer vision tasks that allows users to label and manage large datasets quickly and efficiently. Integrating CVAT with an MLFlow framework can streamline the data labeling process and make it easier to track and analyze the performance of your machine learning models. Here is…
Read more

Dimensionality reduction and how it helps reduce the search space by leveraging known information

Removing dimensions or making invariant features is a technique used to reduce the search space in a problem by eliminating certain variables or making them irrelevant. This can be especially useful in LiDAR slam, which has a 6Dof search space (x, y, z, roll, pitch, yaw). By reducing the dimensions, the number of permutations in…
Read more

Data balancing to remove data bias, do a deep dive on different approaches

Data balancing is the process of ensuring that a machine learning dataset is representative of the real-world population from which it is drawn. This is important because if a dataset is biased, then the machine learning model that is trained on that dataset will also be biased. Bias in machine learning models can lead to…
Read more

How data structures impact time complexity of code

Data structures are the foundation of efficient algorithms and play a crucial role in determining the time complexity of a piece of code. Time complexity refers to the amount of time it takes for an algorithm to complete, and it is a measure of how the runtime of an algorithm grows as the input size…
Read more

Techniques to Boost True Positive Rates using Independent Combinatorics

True positive rates, or the proportion of positive cases that are correctly identified, are an important consideration in many areas. One way to boost true positive rates is to utilize independent combinatorics, a set of techniques that involve combining multiple independent pieces of information or evidence to make a decision. Here are some specific techniques…
Read more

Improving Model Performance from 99.9% to 99.999999%

Artificial intelligence (AI) has come a long way in recent years, with many industries adopting it to improve efficiency and productivity. However, there is always room for improvement, and one area where AI can be further enhanced is in terms of accuracy. Currently, many AI systems have an accuracy rate of around 99.9%, which is…
Read more

Message Queues in Multi-Threaded Applications

Message queues are a software component that allow different parts of a system, or different systems, to communicate with each other by passing messages. They are often used in architectures that are distributed, meaning that they consist of multiple independent systems that need to communicate with each other. One common use case for message queues…
Read more

Wheel Odometer and How it Helps Calibration of Accelerometer

Wheel odometry is a method used to measure the distance traveled by a vehicle by tracking the rotations of its wheels. This can be done through various methods such as using encoders, sensors, or by measuring the speed of the wheels using an OBD (on-board diagnostics) sensor. One of the main applications of wheel odometry…
Read more

ROS 1 vs ROS 2 Tradeoffs and Advantages

ROS (Robot Operating System) is a popular open-source robotics framework that provides libraries and tools for building robot applications. There are currently two versions of ROS: ROS 1 and ROS 2. In this article, we will explore the tradeoffs and advantages of both versions and discuss the tools available for converting between them. We will…
Read more

Block-NeRF and it’s applications in autonomy

Block-NeRF (Scalable Large Scene Neural View Synthesis) is a method for generating 3D reconstructions of large scenes using neural networks. It is based on the concept of neural radiance fields (NeRFs), which are a representation of a scene as a function that maps 3D coordinates to the radiance (brightness) observed at those coordinates. Block-NeRF is…
Read more

Ground Truth Report : Downtown San Francisco

The RoadMentor ground truth report is a detailed analysis of data collected from a drive through downtown San Francisco. The data, which includes base map tiles, ROSBAG files, camera images, LiDAR scans, GPS and IMU data, and fused vehicle trajectory, was collected using a variety of hardware and sensors, including a Ouster OS-2 128 beam…
Read more

Scene Segmentation

The scene segmentation module is used to semantically describe the pixels in image data. It is useful for things like free space detection, object recognition, cross-view localization, image filtering based on label class, compression, etc.


The visualizer allows you to explore out datasets through time and space. You can navigate as a time series. We are adding geospatial indexing soon and the ability to annotate and queries will be added as well.

Object Recognition

The Object Detection pipeline is supported both on the cloud and on the edge using a dockerized container. It assumes you are using an NVIDIA Cuda enabled machine.

SLAM SLAM algorithm without the usage of GPS or an IMU.