Discover How SunTec India is Empowering AI Development with Data Annotation Services

Enterprises that Need Data Labeling Support for AI Development

Tracking Drones and Annotating Aerially Shot Videos for a Tech Company

Our client is a technology company specializing in drone-based aerial surveillance and data collection. They needed labeled videos of drones flying over fields to train their object detection algorithm. However, the footage was shot from other drones using standard and infrared cameras across days and nights, which complicated annotation. Thermal signatures, low-light conditions, and videos where the drone being annotated was too far from the camera presented challenges in labeling.

Additionally, the drones moved unpredictably in various directions—up, down, forward, and backward. This randomness made it difficult for automated tracking algorithms to maintain a consistent lock on the drones' positions. As a result, the video frames needed manual adjustments before they could be annotated.

To address the challenges posed by infrared footage and low-light conditions, we adjusted the opacity of infrared frames to enhance visibility. We made manual adjustments for each frame to compensate for the unpredictable drone movements. This involved manually drawing bounding boxes around the drones and assigning unique identifiers to ensure accurate tracking and annotation.

• Image Annotation Tool : Locally hosted CVAT
• Image Annotation Technique : Bounding box annotation
• Team Size : 20 annotators
• Project Volume : 100,000 frames (55 hours of video)

The SunTec team introduced a 30% improvement in the client's object detection algorithms, significantly enhancing their drone surveillance accuracy. The improved training data also resulted in a 20% increase in operational efficiency.

Annotating Vehicles in Aerial Images for a Government Agency

Before

After

The client is a government agency responsible for urban planning and development. They utilize aerial imagery to monitor and manage city traffic and infrastructure. Our team had to annotate those aerial images, i.e., to identify and categorize approximately eight object classes, including cars, SUVs, vans, pedestrians, motorbikes, cyclists, trucks, and buses.

The images varied in resolution and clarity, presenting challenges in accurately identifying and labeling vehicles, especially in densely populated urban areas and under varying lighting conditions. We used 100% zoom to identify the objects and implemented a multi-pass annotation process, which involved multiple rounds of quality checks and adjustments to maintain precision.

• Image Annotation Tool : Label|mg
• Image Annotation Technique : Bounding box annotation
• Team Size : 5 annotators
• Project Volume : 2000+ images

Our detailed annotation process led to a 35% increase in the accuracy of the agency's traffic analysis models. The enhanced data accuracy also resulted in a 20% improvement in the agency's ability to monitor traffic flow.

Vehicle Damage Detection and Labeling for an Insurance Company

Before

After

Before

After

We collaborated with a leading auto insurance provider in the UK to detect and label vehicle damage in the images submitted for claims (e.g., dent, rust, scratch).

It was challenging to distinguish between dents and reflections caused by shadows. Other issues included identifying dents in darker images and detecting scratches and chips that were not easily visible. Additionally, maintaining consistent output due to variations in image quality posed significant difficulties. We maintained constant feedback loops with the client to overcome these challenges and ensured timely query resolution.

• Image Annotation Tool : Locally hosted CVAT
• Image Annotation Technique : Polygon annotation
• Team Size : 10 annotators
• Project Volume : 3000 images

Our efforts resulted in a 40% improvement in the accuracy of the client's damage detection algorithms and a 30% reduction in claims assessment time.

Image Annotation to Identify Litter on Streets for a Municipal Agency

The client is a municipal government department responsible for city maintenance and public services. They needed our assistance to train a street maintenance tracking system. We had to annotate public bins, potholes, manhole covers, and litter in street images.

The challenges in this project included variations in image quality due to different lighting and weather conditions, interference caused by pedestrians or vehicles, and the diverse appearances of objects in various urban environments. We employed bounding box and segmentation techniques to ensure accurate image labeling. Regular feedback sessions with the client helped refine our annotation process and resolve any queries promptly.

• Image Annotation Tool : Locally hosted CVAT
• Image Annotation Technique : Bounding box annotation
• Team Size : 10 annotators
• Project Volume : 3000 images

Our team’s efforts improved the object detection accuracy of the client’s maintenance tracking system by 45%.

Annotating Water Bodies in Historical Map Images

This client- an academic institution -was involved in studying and preserving natural water bodies through historical data analysis. They were training an AI/ML solution for geographic data mapping. To facilitate automated map analysis, they required accurate annotation of water features in historical map images.

However, the maps varied in quality, and issues like faded details and diverse cartographic styles complicated the annotation process. We employed digital restoration to improve image clarity and used a polygon tool to precisely delineate water bodies in the maps.

• Image Annotation Tool : Locally hosted CVAT
• Image Annotation Technique : Polygon tool (masking)
• Team Size : 3 annotators
• Project Volume : 1200 images

The client noted a 40% improvement in the object detection accuracy of their algorithm, as well as better performance in recognizing and categorizing water bodies.

Text Annotation for a National Restaurant Chain

This client operates across multiple states. They frequently update the menus (containing beverages and food items) of all outlets and need accurate categorization of each item as alcoholic or non-alcoholic for customer clarity and legal compliance. They hired our text annotation team to review a comprehensive list of menu items and label them accordingly.

• Text Annotation Tool : MS Excel
• Text Annotation Technique : Text Classification
• Team Size : 5 annotators
• Project Volume : 50,000 records

We appointed a subject matter expert in the field to lead the project, who guided the rest of the team in research and verification methods to identify whether each item in the list contained any alcohol or not. We utilized online resources, including official menus from the restaurant's website and ingredient lists from reputable sources. A double-check system was also implemented where each item labeled as unclear was reviewed by another annotator to ensure accuracy. We finally delivered a comprehensive, verified dataset to the client, enabling them to confidently update their menus.

Data Annotation Tools

Our Team is Familiar with Prominent Data Annotation Tools

Data Annotation Services

Our Data Annotation Capabilities Include