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

Segmenting and Annotating Images of Human Palms for an Astrology App

The client is a technology company developing an astrology application for palm reading. They were training an AI algorithm to identify palm lines on the hand images uploaded by users, compare them to existing guidelines, and produce a reading. This endeavor required labeled palm images so the model could be trained to identify palm lines.

We had to classify and label diverse areas, including finger points (5 regions), bottom of fingers (5 regions), and hand mounts (9 different regions per palm). However, the images also had unrecognized areas, like the gaps between fingers, which needed to be categorized. Additionally, unclear images affected palm line recognition. Identifying hand mounts became difficult if the images were captured with varying hand postures.

We made manual alterations to accurately identify and categorize the gaps between fingers that the automated tools could not recognize. We used Google images as references of palm lines for precise labeling. Darker or unclear images were processed by adjusting their opacity. We also passed the labeled images through an in-house QC process to ensure all required points across each palm image were correctly annotated.

• Image Annotation Tool : LabelBox
• Image Annotation Technique : Polygon and polyline tools
• Team Size : 10 annotators
• Project Volume : 10,000 images

Our team enabled a 25% increase in the application's accuracy in palm line identification.

Video Annotation to Train a Waste Classification Model

Raw image

Annotated image

This client is a leading educational institution renowned for its research contributions in AI and automation. They needed accurate video annotation services to train an AI-driven waste classification system for improving recycling processes. The project involved analyzing footage from CCTV cameras of waste items on conveyor belts.

We customized CVAT to enforce the client's project-specific labeling instructions, automated frame preprocessing, and developed a custom data export pipeline to append client-specific metadata fields to the final training dataset. This enabled us to categorize each item correctly from a fixed list of 16 waste types.

• Annotation Tool : CVAT
• Video Annotation Technique : Bounding box labeling with human judgment for edge cases

Our efforts resulted in a 98-99% labeling accuracy, improving model precision by 21% and accelerating the annotation process by 42%, helping the client enhance their AI-driven waste classification system.

High-Volume Drone Image Annotation for a Livestock Detection Model

Raw image

Annotated image

This client is a leading AgriTech company that uses drones and AI to deliver precise livestock-counting solutions. They required accurate annotation of aerial images captured over large feedyards and pastures to train their livestock detection AI model.

We faced obstacles such as inconsistent lighting, overlapping cattle, and background clutter, which made it challenging to identify object boundaries. To overcome this, we created clear labeling guidelines, used oval-shaped annotations in QuPath for better object representation, and implemented a multi-level quality assurance process that ensured precision.

• Annotation Tool : QuPath
• Image Annotation Technique : Oval annotations for precise cattle identification
• Team Size : 6 annotators
• Project Volume : 10,000+ aerial images annotated per month

Our efforts achieved 95%+ annotation accuracy, improving the operational efficiency of the client’s livestock detection model.

Boosting Corrosion Detection Accuracy with Image Labeling

Raw image

Annotated image

This client is an infrastructure digitization company that uses AI to help its end users manage and maintain physical assets. They required precise image segmentation to identify and label corrosion on telecommunication towers.

We addressed the challenge of subjectivity in labeling corrosion, as rust appeared differently across materials and lighting conditions. To ensure consistency across multiple annotators, we created a detailed reference guide that standardized how corrosion should be identified.

• Annotation Tool : Label Studio
• Image Annotation Technique : Pixel-level segmentation with smart brush and polygon tools

The project achieved 99% inter-annotator consistency, a 25% improvement in model precision, and 95%+ labeling accuracy, all verified through a multi-stage quality assurance process.

AI for Environmental Monitoring: Satellite Image Segmentation for Accurate Water-Ice Detection

Raw image

Annotated image

This client is a leading European technology company specializing in environmental monitoring and satellite data analysis. They required precise semantic segmentation on RGB satellite images to train an AI model to distinguish between water, solid ice, and slushy ice across seasonal variations.

We faced challenges distinguishing solid ice from slushy ice and maintaining consistency across the images with seasonal variations. To solve this, we worked with subject matter experts to create clear labeling guidelines, trained annotators extensively, and used CVAT, which is exceptionally effective for labeling tasks that require pixel-level precision.

• Annotation Tool : CVAT
• Image Annotation Technique : Semantic segmentation on RGB satellite imagery
• Team Size : 20 image labeling professionals
• Project Volume : 8,500 satellite images

We achieved 98% annotation accuracy and a 99% client acceptance rate, exceeding the minimum accuracy threshold and delivering all datasets two weeks ahead of schedule.

Precise Food Item Labeling to Train AI Agents for Efficient Restaurant Operation Management

Raw image

Annotated image

This client is an AI company specializing in restaurant and food delivery technology. They provide operational solutions (AI agents) to restaurant chains, enabling automated order verification, menu digitization, dispute resolution, and more. To ensure reliable real-world performance for these AI agents, they needed highly accurate image annotations.

We used polygon segmentation to label irregularly shaped food items across a large dataset, ensuring the AI could correctly identify and verify menu items.

• Annotation Tool : CVAT
• Image Annotation Technique : Polygon segmentation with human review
• Team Size : 10 annotators
• Project Volume : 20,000+ images annotated

Our efforts achieved 98% annotation accuracy. Additionally, by developing a detailed labeling taxonomy that covered diverse food categories, packaging types, regional variations, and restaurant chain-based variations in food items, we enabled multi-chain AI agent deployment without the need for client-specific retraining.

Precise Drone Image Annotation for Automated Bird Nest Detection on Power Grids

Raw image

Annotated image

This client is a German technology company specializing in digital data collection and AI-powered object detection for asset-intensive industries. They needed accurate drone image annotation for an automated bird nest detection system to monitor power grid infrastructure and ensure safety.

We faced several challenges, such as varying environmental conditions and the need to classify bird nests in crowded or partially obscured images. To address this, we implemented clear guidelines and used precise bounding box annotations, supported by a multi-tier quality assurance process to ensure consistent and accurate labeling.

• Annotation Tool : Client’s proprietary annotation platform
• Image Annotation Technique : Bounding box annotation
• Project Volume : Over 15,000 aerial images

Our efforts resulted in 95%+ annotation accuracy and we also maintained an error rate below 5%.

Aerial Image Annotation to Train a River Monitoring AI Model

Raw image

Annotated image

This client is a German technology company specializing in AI-powered asset management solutions. They needed high-accuracy image annotation to develop an AI model for detecting river obstructions, ensuring proactive environmental monitoring and effective water resource management.

We faced challenges in distinguishing genuine obstructions from non-obstructive debris and handling varying environmental conditions. To address this, we established clear guidelines and used bounding-box annotations to accurately mark obstructions, enabling the AI system to detect hazards efficiently.

• Annotation Tool : Client’s proprietary annotation platform
• Image Annotation Technique : Bounding box annotation
• Team Size : 6 data annotators
• Project Volume : 75,000 high-resolution aerial images

Our efforts resulted in 98% labeling accuracy, consistently meeting the client’s quality expectations with a revision/rework rate of less than 1%. We delivered 1,500-2,000 labeled images per week, ensuring timely, scalable data annotation.