Welcome to the 4th and final step on how to start using Computer Vision! Previously, we highlighted the benefits of deploying and running a Computer Vision model. We also explained the process of automating human inputs & decisions with our  “Observing>Influencing>Acting” cycle, which we presented in one half of a double feedback loop. In this blog, we will showcase the other half of the loop to help you understand and improve the performance of your Computer Vision model.

Intelligent capabilities

Running algorithms continuously obtain new real-life data and scenarios. By developing them thoroughly using e.g. data augmentation techniques, you can create new data sets that are similar to the training sets. Of course. there will always be changes and new cases that the algorithms have never seen before. But that only highlights the strength of Artificial Intelligence-based Computer Vision solutions even more! Not only are they capable of handling these new, yet unseen, situations, but they can learn, improve, and adapt to changes.

Improve the performance

In our last blog, we explained the process of automating human inputs and decisions using our “Observing>Influencing>Acting” cycle, which is illustrated on the left part of the feedback loop. This left part is handling the visual inputs and observing objects & patterns while sending outputs that influence the behaviors in the real world. To understand and improve the performance of the Computer Vision AI, we will now be looking at the right side of the loop.

  • Performance. After the first deployments of the Computer Vision models, humans are still likely to be in charge of monitoring the models’ performance. This is done by analyzing dashboards and reports on the observations and influencing outputs.
  • Tweaking. A subject-matter expert evaluates whether the desired effects can be managed in the influencing outputs based on the AI-aided observations. After some time, and when the quality of the observations is verified and can be trusted, this human tweaking process can also be automated and improved.
  • Steering. A business expert forms conclusions and makes decisions based on the reports.

When making business decisions, humans often run into limitations e.g. the lack of experience and forecasting capabilities. This is where an AI comes to the rescue to close the loop by learning and improving itself. It will likely detect new patterns that humans can’t immediately see. Continuous production-use data will be captured and used to re-train the Computer Vision model. This way, we can better tailor the AI solution to the latest and ever-increasing production scenarios for specific use cases. This re-training and re-deployment can be done in a controlled way periodically, with extensive re-testing and validations, but also in a continuous way depending on the use case.

Closing the loop

When making business decisions, humans often run into limitations e.g. the lack of experience and forecasting capabilities. This is where AI comes to the rescue, to close the double feedback loop by learning and improving itself. It will likely detect new patterns that humans can’t immediately see. Continuous production-use data will be captured and used to re-train the Computer Vision model. This way, we can better tailor the AI solution to the latest and ever-increasing production scenarios for specific use cases. This re-training and re-deployment can be done in a controlled way periodically, with extensive re-testing and validations, but also in a continuous way depending on the use case.

Conclusions

By having AI that continuously learns and improves, the human actions of “tweaking & steering” can finally be automated. The more data sets based on real-life scenarios you use, the more your AI solution will fine-tune itself automatically. Do you see Computer Vision opportunities in your organization? Our AI Computer Vision platform, AIVI, helps you accelerate the implementation of this continuous “Build > Run > Improve” cycle for your use cases. It is a proven and standardized platform with a lot of benefits to getting started quickly. So, why not give it a try and get in touch with our experts at Mediaan Conclusion?

Read the complete blog series on how to start using Computer Vision here: