Organisations now generate more data than ever, from documents and images to voice recordings and sensor feeds. We are sitting on a veritable goldmine – and unless we can make the gold work for us, it is a wasted resource. The process that makes this data usable for intelligent applications is data annotation.
We have moved on from code-dependent software to tools that mimic human thought and behaviour. From chatbots to facial recognition software to auto - driven cars, the number of applications AI powers is mind-boggling. What these technologies have in common is the use of annotation.
Annotated datasets are crucial in training AI/ML models to perform optimally. Without the proper use of data annotation, a self-driven car would not be able to differentiate between a green traffic light and a pedestrian, and facial recognition software would not be able to tell people apart.
In this article, we look at data annotation use cases by data type and by industry, with practical examples. Data annotation is the labelling of raw data so that AI models can learn from it. Our guide to AI data annotation explains the main types and techniques in detail.
Data annotation use cases by data type
Depending on the type of data being used in a gadget or application, there are five fundamental types of data annotation, as follows, along with their common use cases:
Text annotation
Text is the most utilised data category. Text annotation labels textual data so that the AI model can make sense of text inputs within various contexts. Text annotation includes metadata on entities, relationships, intent, sentiment, and semantics, so the machine can comprehend nuances and respond appropriately.
Customer service bots, and virtual assistants are among the most popular use cases of text annotation. Used across diverse industries such as healthcare, finance, insurance, and telecom, they are trained to comprehend business-specific jargon and human emotions.
Image annotation
Image annotation labels elements within an image by tagging them with simple bounding boxes or advanced segmentation methods. Image annotation ranges from basic image classification (for instance, labelling humans and automobiles) to complex grouping for computer vision (CV) projects. For more examples, see our guide to image annotation use cases.
Image annotation is commonly used in security and surveillance equipment and medical imaging technology. Agriculture is another sector that can drive immense benefits from image annotation through use cases such as crop health monitoring, livestock monitoring, and geo-sensing.
Audio annotation
Audio annotation describes audio data from various sources, such as human speech, animal sounds, musical instruments, and environmental sounds. It includes speech transcription, speaker demographics, semantics, sentiment recognition, and the classification of sounds into identifiable categories such as music and traffic.
Audio annotation is crucial in natural language processing (NLP) applications such as speech-to-text transcription modules, GPS navigation systems, and Cortanavoice assistants such as Alexa and Siri. The use of audio annotation is also popular in the Media and Entertainment industry and online music platforms.
Video annotation
Like image annotation, video annotation is also used to train CV systems. However, unlike image annotation, video clips are annotated on a frame-by-frame basis, as the object of annotation is in motion.
Video annotation is extensively used in retail and traffic surveillance systems for object detection, movement tracking, and live location tracking. An advanced use case is autonomous vehicles that use CV to navigate roads and identify obstacles on the street.
LiDAR annotation
LiDAR (Light Detection and Ranging) annotation is a complex image labelling process that enables point cloud processing. It uses multiple sensors to calculate dimensions such as depth and the distance between two objects.
LiDAR data is commonly used to compute geospatial measurements and generate 3D maps. The automobile industry extensively uses LiDAR annotation in advanced driver assistance systems (ADAS) to improve road safety.
Applications of data annotation by industry
The use of annotation powers several industry-specific technologies. With AI/ML tools growing in complexity, applications of annotation services in modern businesses usually combine different annotation types.
Here is how various industries are driving growth through the use of annotation:
Retail
The applications of annotation services are increasingly enhancing efficiency and customer satisfaction in the retail sector. The use of annotation powers multiple processes in retail, such as inventory management, process automation, after-sales service, and visual search enhancement. Common examples include shelf and planogram monitoring, product tagging for search, and in-store shopper analytics. For instance, Infosys BPM annotated 70,000 images and 5,000 videos with over 98% accuracy to support inventory management for a major US warehouse retailer.
E-commerce
Annotation-based technologies such as chatbots, visual search, voice search, and personalised product recommendations are popular with e-commerce sites. Typical examples include tagging product attributes such as colour, size and material so shoppers can filter results, and labelling customer queries so chatbots route them correctly.
Automotive and autonomous vehicles
Autonomous and assisted driving is one of the most annotation-intensive applications. Advanced driver assistance systems rely on labelled camera images, video and LiDAR point clouds to recognise lanes, vehicles, pedestrians and traffic signs, and to judge distance in real time. Every driving scenario, from night driving to heavy rain, needs its own annotated examples. Infosys BPM supports this work with HD image and point cloud annotation for a global technology company.
Healthcare
Healthcare is another sector that extensively uses annotation in digital imaging technologies like X-Ray, MRIs, CT scans, and PET scans. Annotated technologies also aid diagnostics, genetics research, and drug discovery. Medical annotation usually needs clinical experts: in one project, Infosys BPM deployed licensed dental experts to annotate data for a healthcare client’s dental AI model, completing 99% of annotation tasks on time.
Education
The education sector uses annotation-powered processes such as OCR (Optical Character Recognition) and NLP (Natural Language Processing) to devise online tests, enable remote learning, and transcend linguistic barriers in teaching.
How can Infosys BPM help?
Data annotation outsourcing is crucial for modern businesses that aim to enhance their impact through AI/ML models. Infosys BPM provides reliable annotation service, helping clients across various industries to build high-quality training data at scale. Choosing Infosys BPM as your data annotation outsourcing partner helps your in-house teams to save time, effort, and resources. Its managed annotation teams deliver more than 98% accuracy and handle more than 150 million image annotations a month.
Frequently asked questions
Industries with large volumes of data or specialised labelling needs outsource most often: automotive and autonomous driving, healthcare, retail and e-commerce, technology and financial services. Outsourcing gives them trained annotators, domain experts and quality processes without building large in-house teams.
It depends on the task and how varied real-world data is. Most teams start with a pilot dataset, measure model accuracy, and then add annotated examples where the model struggles, especially rare or edge-case scenarios.
Sometimes. Raw data can be reused, but labels are usually specific to the task, so a dataset annotated for object detection may need new or extra labels before it can train a segmentation or classification model.
Inaccurate or inconsistent labels lead to models that make wrong or biased predictions, need extra training cycles, and can create safety or compliance problems in areas such as healthcare and autonomous driving.
