2D BOUNDING BOXES
This is the simplest type of data labeling that outlines relevant objects to describe a relatively precise location of an object.
3D BOUNDING BOXES
These data annotations help algorithms understand the orientation and relative structure of an object.
SEGMENTATION MASKS
These data annotations provide the finest level of outlining for objects, and are typically used when precision is required at all costs.
DEPTH MASKS
Depth annotations describe the distance of an object relative to other objects.
SURFACE NORMALS
Normals convey information about surface direction and detail.
KEYPOINT/POSE
This helps machine learning algorithms understand the pose of a person or object.
FREQUENTLY ASKED QUESTIONS ABOUT DATA LABELING SERVICES
What Does Data Labeling Mean?
In machine learning, data labeling is the process of identifying raw data (images, text files, videos, etc.) and adding one or more meaningful and informative labels to provide context so that a machine learning model can learn from it.
What is the Difference Between Labeled and Unlabeled Data?
Labeled data is data that comes with a tag, like a name, a type, or a number. Unlabeled data is data that comes with no tag.
What is Metadata?
In machine learning, algorithms are trained from annotated datasets based on images or, in our case with synthetic data, 3D environments. Raw data doesn’t tell us — or machines — much, but it’s the metadata, via annotations, that combine together to create predictive algorithms for computer vision.
What are Annotations?
Annotations describe the data. For example, if you’re training a deep learning algorithm to recognize the difference between certain animals, annotations could be a list of values that represent the locations within an image or environment of dogs and cats. Annotations may also include their orientation — which way they’re facing — their color, even their age.
Eventually, once an algorithm has ingested enough metadata — annotations — it can then recognize cats versus dogs and, eventually, generalize to new and unseen cases — breeds, skinny versus overweight, even mood — as it’s given more annotations to learn from.
Why are Data Labeling Services Important?
The labels used to identify training data must be informative, discriminating and independent to produce a quality machine learning algorithm. A properly labeled dataset provides a ground truth that the ML model used to check its predictions for accuracy and to continue refining its algorithm.
Difficult to reach places such as underwater locations, hard to replicate scenarios like extreme weather, varying perspectives from top down to bottoms up, costly real world tests like crashes all become easily simulated in virtual worlds through data labeling.