At first it sounds good, but when I dived into it I don't see point of it in the current app.
Let me explain.
We are using const number of last labels (inferences) to increase accuracy of model. One inference is one camera frame. So if we limit number of frames than it will take longer to recognize banknote.
On older devices it is slower, on newer it is faster.
What's the place of a battery consumption in whole that discussion? When the user has better phone it will take him less time to get the recognition thus he will quicker leave the app making the consumption lower. I think it's fair. We can't jump above the performance of the smartphone.
At first it sounds good, but when I dived into it I don't see point of it in the current app. Let me explain. We are using const number of last labels (inferences) to increase accuracy of model. One inference is one camera frame. So if we limit number of frames than it will take longer to recognize banknote. On older devices it is slower, on newer it is faster. What's the place of a battery consumption in whole that discussion? When the user has better phone it will take him less time to get the recognition thus he will quicker leave the app making the consumption lower. I think it's fair. We can't jump above the performance of the smartphone.
Let me know what you think!