FAST AND RELIABLE POINT CLOUD ANNOTATION
- We offer solutions for all the steps in the AI development life cycle from dataset generation, dataset annotation, to developing ML models
- Working with lidar and radar data for automotive applications is our area of expertise
- High-quality dataset generation coupled with automotive annotation tools we have developed and use will accelerate your development and aid you in bringing a quality product to the market
- We know how to deliver a stable platform for the visualization of any 3D point cloud data into 2D space
Why Automotive Annotation and Why NOVELIC
Fast and reliable annotation represents one of key challenges for the development of autonomous driving.
This field can still be seen as the one in the early stages of development, however the NOVELIC team has already gained a unique experience from the most advanced projects.
1. First You Need a Dataset
Lidar, Camera and Radar Recording
With our road scanning lidar/radar setup, extensive data acquisition is possible while simultaneously being optimized for sensor fusion so that a high-quality dataset generation is guaranteed.
Real-World Traffic Scenarios
Lidar and radar mounted vehicle for the recording of roadside and traffic scenarios in the European urban, rural, or highway environments.
Synthetic Data Generation
Low-cost synthetic data generated using an advanced 3D engine for scenarios when a real-world recording is impractical or unsafe, coupled with data capture modeled after real sensors.
2. Accelerated and Precise Data Annotation
Multi-Sensor Automotive Annotation
Lidar, radar, and image annotation from within a single tool, supporting a keyframable timeline for quick interpolation of annotations between frames.
Effortless Coordination of Annotation Tasks
With our DevOps-inspired web platform for task tracking, you will have easy access to powerful task tracking features and contact with delegated annotators. Defining annotation tasks, annotation labels, and their attributes are made simple with our web platform, and will, no doubt, feel familiar with other task tracking solutions you may have used.
Custom-Made, with Speed and Accuracy in Mind
Speed and accuracy are the main parameters for optimization when it comes to annotating large collections of data. This is why speed and accuracy were our top priorities when designing the annotation tool. Every extra click we save per annotation sums up to hours and even days of labor saved per dataset. This tool features semi-automatic (AI aided) single-click annotations, with a shortcut-rich and user-friendly interface.
Delivering annotated data can be done in customer-defined formats, as well as standardized formats such as CVAT XML or our custom JSON files.
Inspired by Industry Standards
Deliberately inspired by industry-leading standards for automotive annotation tools and their workflow (such as CVAT).
Web Task Management Portal
All Sensor Annotation from the Single Tool
NOVELIC Workflow Example
3. Development and optimization of AI
Deep Learning Powered Classification
Classification of custom-defined 3D objects in point cloud space by utilizing the latest cutting-edge research in deep learning models for point cloud data.
Adapting and Preparing Dataset
A thorough analysis of the dataset to ensure consistency so that data will be utilized properly in the training phase.
Training, Validating, and Testing Cycles
We use separate training, validation, and testing datasets during the learning process to fit the parameters of the classifier, tune hyperparameters and ensure the quality of the trained model by eliminating common problems such as overfitting.
Quick Inference Fit for Any Use Case
Working on performance optimizations and reducing the inference time of implemented models to achieve a more flexible product, fit for any hardware or time constraint.
Brief Overview of NOVELIC Automotive Annotation Services:
ANNOTATION OF ANY CAMERA, LIDAR, OR RADAR DATA
When you deliver datasets to us through our easy-to-use task management platform, you get fully labeled surroundings in the shortest period.
We develop new categories of algorithms with great confidence. We help you to improve and reconfigure your hardware and feed our machine learning model delivering highly reliable data. There are two ways we can execute algorithm development:
a. Full-fledged development
b. Existing algorithm optimization, fine-tuning, and productization.
UNIQUE PLATFORM FOR ANNOTATION COLLABORATION
We have developed a full customer management portal for annotations as well – this task management environment is easy to use. It does not require the use of different crowd-working solutions for creating annotations, which enables source data hosting on customers´ servers.
YOUR TECHNOLOGY NEED EVOLVES INTO END PRODUCTS
We can lead all the phases from the initial software prototype development & evaluation to the fine-tuning with product development.