Annotation Services to USA Artificial Intelligence (AI) Industry

Introduction

High performance and reliability of Artificial Intelligence (AI) models depend on high volumes of accurately annotated data. One of the top AI solutions providers in the USA needed to connect with a dependable annotation partner for their AI machine learning and computer vision and data analytics projects. The client wanted a solution that could accommodate a large volume of annotations, while ensuring quality, reliability and timely project delivery.
Infosearch BPO was chosen to offer full annotation services, giving the client the ability to speed up the process of AI model development and quality of training data.

Background of the Company

The client is an Artificial Intelligence company based in the USA, that specializes in AI powered solutions for a variety of industry sectors such as smart automation, healthcare, retail and logistics. The organization applies machine learning and computer vision technologies to create intelligent systems that allow it to detect objects, recognize images, analyse patterns, and make decisions automatically.
The need for precise labelling of the datasets grew as the company applied more and more of its AI projects. The client wanted to have an experienced client partner that can provide them with quality data annotation services for various project types.

Objectives

The main goals of the project were:

• To offer precise and uniform image and video annotations for training machine learning models.
• To meet the need for large-scale annotation needs and scale up resources to accommodate growing project size.
• To achieve high accuracy of the annotations with robust QA process.
• To provide annotated data on time to support the client's AI development needs.
• To set up rules for annotation in order to ensure consistency between data sets.

Process

Infosearch BPO adopted a structured annotation workflow to guarantee accuracy, consistency and efficiency during the project.

Requirement Analysis

The project team collaborated closely with the client to gain a grasp of the requirements for annotation specifications, labeling guidelines, data formats, quality standards and expectations for delivery.
Students will learn about the importance of forming teams and training them.
A team of trained annotators and quality analysts was set up for the project. Team members were extensively trained on client specific guidelines and project requirements of annotation.

Data Annotation

The following activities were carried out in the context of the annotation activities:

• Bounding box labelling for object detection.
• Precise object segmentation by polygon annotation.
• Semantic segmentation
• The categorization and classification of the images.
• Video frame-by-frame annotation
• Use attribute tagging and metadata labeling.

Data Annotation

A multi-level quality review process was set-up:

• For initial annotation, the text is annotated by trained annotators.
• It is reviewed by senior quality analysts on a secondary basis.
• Random sampling for quality validation
• Continuous feedback and correction mechanism(s)

Data Delivery

The datasets were handed over with annotations on time through secure transfer channels and progress reports were furnished to the client regularly.

The data volume is high.

This project required processing of large amounts of images and videos in limited time. Effective resource utilization and project management were crucial for productivity.

Annotation Consistency

Consistency between different annotators was achieved by providing detailed guidelines and ongoing monitoring, with a focus on avoiding variations in labeling.

Complex Object Identification

Some datasets had multiple objects, poor image quality and complex scenes, which added to the complexity of the annotation and review process.

Evolving Project Requirements

The client continued to develop their own AI models, which necessitated frequent changes to the annotation specifications, and the necessary adaptation of the annotation team.

Results

The project resulted in a number of important benefits for the client:

• To be able to annotate large scale images and videos successfully within agreed timelines.
• Collected accurate annotations with a quality assurance process.
• Enhanced the quality of training sets for building machine learning models.
• Supported accelerated AI model development and testing cycles.
• Proposed and rolled out scalable AI annotation resources to support expansion of AI projects for the client.
• Built a long-term relationship based on high standards, availability and on-time project completion.

The project proved to be a success, which enabled the client to speed up their efforts in developing AI and minimise the running costs of in-house annotation management.