Existing solutions to image dataset classification are highly monotonous, and require significant human resources, therefore are time consuming and expensive.

We aimed to provide a solution that leverages AI to take the monotonous burden of annotation off human workers, and to make the process quicker and more efficient.

To be accessible to everyone, the tool needed to perform well on commodity hardware, running in the browser. We used a special technology stack to create a solution that is able to handle millions of images, without sacraficing performance.

We started the research phase with competition analysis. We conducted usability tests on multiple competitor solutions, identified pain points and areas for improvement. Based on these insights, we developed several prototypes for our solution over multiple iterations.

During development, a specialized tech stack was used, centered around Python on the backend, and WebGL on the frontend, to achieve a very performant solution capable of handling millions of images.

\Spective outperformed all competing solutions in image classification across multiple test datasets during our performance evaluation.

The solution successfully reduced annotation time by 80%, and increased label accuracy by 10% on average. Users reported of a more engaging and less monotonous annotation process, and preferred \Spective over other solutions.

Thanks to the specialized technology stack, \Spective handles big datasets with ease, providing great performance even on mobile devices.