Computer vision is one of the most disruptive technologies of the recent decade. Whether you're thinking about consumer technologies such as autonomous vehicles and Face ID, or something that works more behind the scenes like AI-powered geospatial intelligence and medical imaging, computer vision brings new opportunities across industries.

Breakthroughs rarely come easy, and that is the case with practical applications of computer vision as well. Today, developing a computer vision model requires a tremendous amount of labeled images and enormous computational power.

• Data labeling: Most methods of labeling images today are labor-intensive and can require entire teams of data labelers to produce a sufficient dataset.

• Model training (and retraining): Utilizing powerful cloud GPUs doesn't simply come with increased compute cost but also increases overhead for each data scientist as they need to be supported by DevOps engineers.

These two requirements usually take a massive investment upfront to develop computer vision systems, thus drastically stretching the time to market. Teams that are in a hurry or working with limited resources need to look for creative solutions.

Two Platforms, One Solution

Superb AI and Valohai have partnered to provide an end-to-end solution that tackles these two challenges. Our focus is to make an infrastructure solution that makes computer vision possible with a timeframe and team structure that is more akin to classical ML.

The two platforms cover the data preparation and the model operationalization aspects, respectively, combined as a continuous pipeline. As production models are never one-offs but rather evolving products, any changes such as new data inputs or code modifications will automatically trigger the pipeline from data preparation to model development to model deployment.

Superb AI Suite Platform: From input data to labeled data

Superb AI has introduced a revolutionary way for ML teams to drastically decrease the time it takes to deliver high-quality training datasets. Instead of relying on human labelers for a majority of the data preparation workflow, teams can now implement a much more time- and cost-efficient pipeline with the Superb AI Suite.

A typical data preparation pipeline might contain the following steps:

• Data ingestion: Input data (images and videos) are often extracted from various sources. Users can upload input data into Superb Suite as raw files, via the cloud (AWS S3), or via Superb Suite's SDK/API.

• Ground-truth data creation: Having a small amount of initial ground-truth data (data with correct labels) is crucial to kickstart the labeling process. Users can create these ground-truth samples using Superb's built-in simple annotation tool with filtering capability. Superb Suite supports classification, detection, and segmentation tasks for bounding boxes, polylines, polygons, and key-points for both images and videos.

• Automatic labeling: Superb AI's customizable auto-label technology uses a unique mixture of transfer learning, few-shot learning, and autoML - allowing the model to quickly achieve high levels of efficiency with small customer-proprietary datasets. And because the custom auto-label has broad applications, it can be used to swiftly jump-start any project, whether that be labeling your initial dataset for training or labeling your edge cases for retraining. This will drastically reduce the time it takes to prepare and deliver datasets.

• Labeled data delivery: The review and audit process of data labels is vital for the overall quality of the dataset, but in reality, it is almost impossible to review every label manually. Superb AI Suite streamlines the review process by taking advantage of the label accuracy measures estimated by multiple machine learning models. After passing through this rigorous quality control process, the final labeled data is delivered to the MLOps pipeline.

Valohai MLOps Platform: From labeled data to a deployed model

The labeled dataset from Superb AI gets ingested into a machine learning pipeline. Valohai makes it possible to build pipelines with any number of steps run in parallel or sequence. Each step can contain any language or framework and run on specific hardware suited for the purpose  -- all without any engineering overhead.

A typical machine learning pipeline might contain the following steps:

• Data augmentation: Labeled datasets often go through data augmentation, where the original data is duplicated and modified to produce slight variations, such as added blur, discoloration, or changed orientation. This helps combat overfitting in model training. Code agnosticism makes Valohai ideal for data augmentation as it can run any kind of augmentation code.

• Model training: Training is the central piece of the machine learning pipeline. Valohai makes it easy to run model training on any type of cloud or on-premise environment. The platform spins up and shuts down instances automatically so compute time is never wasted.

• Model evaluation: Pipelines usually have an evaluation step to ensure the quality of the model. This step codifies any quality metrics and can, for example, contain comparisons to previous versions of the model. Valohai allows this easily as all the previously trained models are stored on the platform.

• Model deployment: As a final step in the pipeline, the model can be deployed. Valohai can deploy the model to a managed Kubernetes cluster (for online inference) or push it to other systems (e.g., for edge inference).

Conclusion

Practical computer vision applications are still waiting to be unleashed due to the demanding nature of developing them. The demand for human and machine resources grows the project's complexity and stretches timelines and budgets that can be out of reach for many organizations.

However, as shown above, intelligent platform choices make computer vision much more attainable. With Superb AI's automatic labeling technology and Valohai's managed training infrastructure, you can drastically reduce the need for the labelers to get the data right and for the engineers to build and maintain a working infrastructure.

If you are looking for a complete solution for developing computer vision applications, let's schedule a demo and see whether our solution is right for you.

Superb AI is an enterprise-level training data platform that is reinventing the way ML teams manage and deliver training data within organizations. Launched in 2018, the Superb AI Suite provides a unique blend of automation, collaboration and plug-and-play modularity, helping teams drastically reduce the time it takes to prepare high quality training datasets. For more information, please reach out to our team. 

Valohai is the only MLOps platform that automates everything from data extraction to model deployment. Valohai is all about taking away the not-so-fun parts of machine learning. Managing cloud instances and writing glue code is neither valuable nor fun. Our platform does that for you. We're trusted by companies such as Twitter, JFrog, Konux and Preligens. For more information, please reach out to us.

Both Superb AI and Valohai are part of the AI Infrastructure Alliance and dedicated to building the foundation of Artificial Intelligence applications of today and tomorrow.