Ever wondered how important frames per second (FPS) is in video processing for computer vision? In this article, we’ll take a dive into the world of FPS with Carter Brown, a senior computer vision engineer at Kibsi, who knows a thing or two about making computer vision work at breakneck speeds (and cracking a joke or two in the process). Carter will explain why high FPS is crucial for complex computer vision requirements, especially in real-time business applications. Check out Carter’s video explanation below and watch a sample video showcasing Kibsi’s impressive tracking capabilities.
[Placeholder for Carter’s video]
Frames per second (FPS) is a vital aspect of real-time computer vision applications in any business. In essence, FPS refers to the number of images shown to the model per second and the number of images processed end-to-end in the entire Kibsi computer vision stack. But, as Carter likes to joke, “the only thing worse than watching paint dry is watching it at 10 FPS!”
Carter explains that, in many cases, CCTV cameras operate at 10 FPS, which is sufficient for most scenarios. However, when it comes to packing items in a box quickly, as demonstrated in the sample video below, 10 FPS simply won’t cut it. It’s like trying to enjoy a movie in slow motion – not exactly ideal! A worker might grab an object off the shelf, and a tenth of a second later, it’s already in the box. To accurately capture and process this action, we need to up the speed and look at a full 30 frames.
To achieve this higher FPS, Kibsi employs smaller models capable of keeping up with real-time video processing requirements. Kibsi also utilizes a standard tool called TensorRT to make the entire process run faster. Although the technical details of TensorRT may not be crucial here, the point is that Kibsi supports a wide range of model sizes and compilation procedures. This flexibility allows Kibsi to explore various trade-offs to find the most suitable model for each specific task or customer requirement.
Once Kibsi evaluates the trade space and selects the best model, it is uploaded to the cloud, enabling the entire pipeline to run end-to-end. This process ensures that Kibsi can capture every frame and track objects as they enter the box without missing any frames in between. Or, as we like to joke, “Kibsi models are so fast, they’d leave the Flash in the dust!”
Kibsi’s flexibility, combined with its ability to process high FPS, makes it the perfect solution for businesses that require computer vision to operate at high speeds. By offering various combinations of models and compilation strategies, Kibsi can accommodate the unique needs of any business, ensuring smooth and efficient operations.
So, can your computer vision keep up with the speed of your business? With Kibsi, the answer is a resounding “yes.” In fact, it’s so fast you might say it’s almost a whole other FPS (Freakishly Powerful Speed).