Real-time cost · Edge AI
Real time has a cost: when not to compute
A sensor on a mast keeps track of an object in the sky. The heavy model runs on one frame, and the frames in between are carried forward by a far cheaper prediction. Continuous awareness, less unnecessary compute.
Chapters
CONCEPT FILM
The scenes, vehicles and sensors, and the boxes, tracks and labels on screen, are Spikedge's own 3D animation, not detector or controller output. The film makes no measurement claims.
01What the film shows
The interesting question is not whether a model can find the object. It is what the system spends to keep finding it, thirty times a second, for as long as the object is there. Between two frames very little changes, so running detection and classification on every frame pays the same price for information the system already has.
The architecture in the film detects when detection is needed, carries what is already known, predicts what comes next and corrects when the next detection arrives. It is a budget decision made with the deadline in hand: how long the control loop can wait, how far the object travels inside that window and how wrong the prediction may be before the next detection has to land.
The idea comes from published work co-authored by Spikedge's founder: Çalışkan and Koruk, ICHORA 2025 (DOI 10.1109/ICHORA65333.2025.11017067), where detection and classification ran on one frame in four and a dynamic linear regression model predicted the remaining three. That result belongs to that hardware and that task; the film illustrates the principle and does not repeat the measurement.
02Transcript
In the real world, intelligence isn't measured only by accuracy.
It's measured in time.
Processing everything, every time, has a cost.
What if intelligence didn't mean doing more, but knowing when to compute?
Detect when detection is needed.
Carry what is already known.
Predict what comes next.
Because in real systems, the best decision is often the one made before the moment arrives.
