DATA ANNOTATION SERVICES

Most of us have probably heard of Artificial Intelligence (AI) and Machine Learning (ML). Many of us would have experienced AI at work in one way or another.

• A robot moving through a shop floor and navigating around objects in its path
• Detecting a malignant tumor in a pathology film
• A Resume parser distributing an unformatted Resume into discrete data fields

Some would recognise it as the work of AI and some would not.

Fewer still might appreciate the process that is behind making machines (or software programs) acquire human-like intelligence and act in accordance.

This might disappoint some, but AI in machines is neither the result of the discovery of a rare mineral from the Earth’s core nor a hitherto unrecognized magnetic wave, it is the result of painstaking training and programming done by the human hand and human intelligence.

The process of training a computer to pick up signals from data is known as Machine Learning (ML). Machines are not able to interpret data in the same manner as human beings. For example, when a human being sees a smiling face, she will instantly recognise it as one. For a computer, it is just data at the start and has no meaning. It would need to be fed to the machine in a manner that it can understand and associate it with meaning and action. For example, one way could be to teach the machine to identify a smile from the shape of the lips. If the middle point is below the imaginary line connecting the ends of the lips, it is a smile.

This is the process of image, video and data annotation services, handled by data annotation companies. Of enriching the raw data so that a computer can associate some meaning to it. And doing it again and again, allowing the computer program to draw inferences and ‘learn’ so that when the training stops and it is presented with the next image, it can now interpret it as a smiling face or a frowning face. The magic behind AI lies in the quality of data annotation.

Of course, this is a simplistic example for illustration purposes.

Types of Data Annotation 

There are many ways in which data can be annotated by a data annotation company for ingestion by a software program through ‘computer vision.’ The usage of a particular method will depend on the type of data and the objective of the exercise. A few common types are detailed here:

Bounding boxes

The primary use of bounding boxes is in identifying the location of an object and its size. It is done by drawing a rectangle tightly around the object, in a way ‘binding’ the object with the rectangular box. The box, thus, becomes the ‘bounding box.’ It is a two-dimensional representation and can be easily identified by x and y coordinates.

Lines and splines

A line is a line as we all know. But what is a spline? While there are perhaps complex mathematical definitions that are more accurate, a spline is essentially a flexible (not straight, not circular, not curved, but just flexible) that represents a ‘best fit’ across multiple points. Lines and splines are used in lane detection and identification, which is useful for autonomous vehicles, a rapidly evolving application of AI.

Semantic segmentation

This type of image annotation is done at a pixel level in an image. Each pixel in the frame is attributed to one and only one of the classes of objects that need to be identified based on the context. In a two-dimensional image, the attribution will only be to the object type visible and on top.  

Entity annotation

Used for textual data, entity annotation identifies named entities. It is used for many applications in NLP (natural language processing) like searching and summarizing.

3D cuboids

This could be visualized as a Bounding Box but with a third dimension, depth. Once again, driverless cars could be an example of the application of this technique of annotation where an estimate of distance could be established. With the third dimension, it would also become possible to estimate the volume of objects if required.

Polygonal segmentation

Using a Bounding Box as the basic object identifier, one could visualize the rectangular box to evolve into a polygon, in other words, a multi-sided figure, that closely fits the contours of the object identified much more than a Bounding Box would.

Landmark and key-point

The basic technique of image annotation  lies in creating a map of the image at hand by creating dots across it and assigning meaning based on the combinations and relativity so created. Used in applications like recognition of faces, interpretation of the expression on the face and land-use pattern (based on aerial images).

Choosing a data annotation company

Choosing a partner for one’s data annotation work is an important step. In any B2B relationship, one partner’s performance impacts on the other. The larger the volumes, the greater the stake, which is why companies carefully evaluate interested parties before taking on one out of the many interested data annotation companies.

Given below is a set of parameters that outsourcers should use to separate suitable providers from the unsuitable ones and select the most suitable one.

What prior experience do they possess in this field?

While it is not recommended that bidders without relevant prior experience be eliminated without consideration, prior experience of providing the same or similar services is of help in enabling the proposed engagement to get off to a flying start, considering that a BPO engagement has to necessarily go through an initial cycle of hiring and training and then ramping up, before efficiency can be achieved.

In fact, taking on an inexperienced partner may fetch good value on other parameters like pricing. For the sake of earning experience in a new stream of work, the vendor may be willing to offer favorable rates as they know they will otherwise not stand a chance vis a vis other bidders. So, place value on prior experience, but don’t outright reject its absence.

Having successfully executed a wide variety of data annotation projects over the last eight years for a wide variety of clients from across the globe, oWorkers is in a position to help your project get off to a quick start and reach optimum productivity in a defined timeframe.

Do they have the Project Management expertise for a smooth transition?

Experience and quality and accuracy are factors that come into play only if the transitioning of the work from client ownership to vendor ownership has been done effectively by the data annotation services. Very often parts of the process slip through the cracks in this ‘lift and shift’ operation from client to vendor. The skill of Transitioning under the overall umbrella of Project Management is an important discipline in the armoury of BPO companies that needs to be deployed again and again, each time they win new business.

The oWorker philosophy is that for a client, the delivery experience starts as soon as the contract is signed. Hence, project management and transitioning are essential components that contribute to the oWorker experience for a client. A key contributor to oWorkers having transitioned and served hundreds of satisfied clients over eight years has been our project management and transition discipline. And we are fine-tuning it with each new contract.

Will they be able to deliver the accuracy required by our project?

Eventually it is about quality and accuracy. There is a reason a piece of work needs to be done. It will perhaps create an input into the next process. The next process requires it to be of a certain quality and accuracy. Else, we all know that the GIGO (Garbage In Garbage Out) principle takes over. If it needs doing, it needs doing well, to the level specified based on the requirements of our process. If there is doubt on this aspect, it might be best to stop the conversation right now and save everyone some time and heartache.

The challenge is that quality, or accuracy, on a project cannot be evaluated unless it has been done. And it will only be done after we have evaluated and awarded them the contract. So, how does one evaluate potential to deliver?

Existing client testimonials are perhaps a good way of establishing commitment to delivering good quality. If they have done it for one or more clients, chances are they value delivery of a high standard and will maintain it for you.

oWorkers has consistently delivered at over 98% accuracy level, which is higher than the level demanded by most client processes. Our QC (Quality Control) and QA (Quality Assurance) processes and an independent team reporting directly to senior management, keep our delivery teams on their toes and ensure that the quality reaching the client is better than their expectation.

Can they deliver fast? Overnight?

Speed is always important in business. Of course, not at the expense of quality. But, a business wants everything. Quality. And speed. Speed allows you to do more, faster. It gives you a competitive edge. A data annotation company doing work for you needs to fit into your scheme of things and deliver such that your process keeps moving at the optimum speed. Even if they take time in establishing a new process, once ramped up and running, it needs to be run efficiently.

oWorkers data annotation services deliver fast. It allows us to do more for you, which allows us to earn more. We deliver overnight on many of our contracts. Our centers are geared to operate 24×7. In addition, we have centers in three of the most desirable locations for BPO services, which, in some cases, provide a time difference with our clients that again facilitate overnight processing. The input we receive is processed and ready by the time the client starts work the next morning.

Is the offered pricing competitive?

This is obviously an essential feature of any commercial engagement, though it is recommended that this be evaluated in the end, after the other aspects have been. The reason is that pricing often overshadows other, equally important criteria and decisions can be taken solely on the basis of the price being lowest.

It must be highlighted that while a low price is always welcome to an outsourcer, an inordinately low number, way off other bidders, should ring alarm bells and not be accepted with open arms. The possibilities could be:

• The vendor is desperate for business at any cost
• The vendor is trying to prevent a competitor from getting the deal at any cost

The risk for you is that if the deal is financially unsustainable for the vendor, they will either withdraw at some stage or cut corners leading to quality issues.

At oWorkers, with our nearshore and offshore centers, you have the potential to save up to 80% on your cost, as confirmed by several of our existing clients. This can translate to upto 3 USD per hour for scale. Our pricing being transparent, we offer you a choice of dollars per unit of time or dollars per unit of work. You choose.

Will they support multilingual delivery?

Data annotation companies ought to be aware that our world is more global and connected today than any point of time in human history. One of the direct fallouts is that companies that do business globally need to provide services in many languages and different cultures. Whether you are already global or aspire to go global, all businesses are trying to expand and either become global, or more global. When you take on a partner, it would be best to take on one who can support you irrespective of the region your work is in, which means the ability to provide support in multiple languages and awareness of cultural sensitivities.

Our language capability is available when you outsource data annotation projects to us. oWorkers currently supports over 22 common languages across our three global centers for a wide variety of data, video and image annotation services, and are willing to add on more if there is a specific requirement.

Do they use current technology? Is our data secure with them?

The backbone of most BPO processes is technology. In fact, the reason for data annotation being a requirement is again technology, Artificial Intelligence. The vendor should be able to demonstrate the commitment to leveraging the power of advanced technologies that will facilitate the task they seek to take on.

Data security is closely linked to technology, as the data worked on is almost always in digital format that gets shared across networks and devices to move the process forward.

Through its partnership with leading providers of data annotation tools, oWorkers leverages the best technology solutions for client projects. We are ISO (27001 and 9001) certified and our employees sign and operate under an NDA and work with strict security protocols. We also offer physical segregation of client processes, where required.

Can they hire in right numbers, when needed? Can they handle spikes?

Along with technology, people, or human resources, are the essential pre-requisites for all data annotation companies. After all, it is human intelligence that will create the training data-sets that will be used to train machines. The partner needs to be able to hire the right resources, at the right price, as and when required by the business, which means occasionally scaling up and down to meet seasonality and other spikes.

oWorkers can ramp up and ramp down very fast. To be more specific, by a hundred headcount in 48 hours. Our deep connections with the local communities and our position as a leading employer gives us access to a perennial supply of interested candidates seeking a job. It helps that we are known to hire people and not use them as contractors or freelancers. This also gives us flexibility in deployment.

Is the company’s financial health conducive to continuity and support?

If financials of a data annotation company present a less than robust picture, there could be a tendency to take shortcuts in delivery, which could adversely impact client deliverables. Also, there is a possibility that adverse financials may force the vendor to take hard decisions like closing down parts or all of the business, which means you will need to go through the vendor identification process all over again.

oWorkers is registered locally in all the global centers it operates from and has been a consistently profitable enterprise. It also pays local and social taxes for its employees and is deeply rooted in the communities it operates in.

Is the management committed to our business?

Eventually the leadership of the data annotation company will decide how important each contract and each relationship is for them. While there are no clear parameters to establish this, it is hoped that an outsourcer’s leadership team will have the foresight to judge this during the several interactions they are likely to have with the partner’s leadership team, to form an opinion.

The Oworkers Advantage

Oworkers is a specialist in Data and Content services for whom GDPR compliance is a requirement, not a choice. Our delivery centers are located in three global locations providing the benefit of business contingency in times of need. We provide the flexibility of working from home to our workforce during these times of the pandemic. We have been a trusted partner of several UNICORN marketplaces over the years.

oWorkers is a data annotation company that strives to create positive social and economic change through employment in underserved communities. By working with us, you help bring motivated individuals into the global digital economy.