I am working on a multiplatform app for object detection applying machine learning, I am using the TensorFlow framework. I notice that in the latest Felgo version the performance of the Android app has considerably decreased. The inference time, the time for processing an image, has increased a lot on Android with the same device, Android version, neural network, TensorFlow version, and app code. I am only measuring the inference time, not the time spend in converting or drawing on the image. These are some estimated inference times on several platforms.
Linux: 100 msec – 200 msec (Laptop). https://drive.google.com/open?id=1vZBKGR-LkH4KLivZWpYapGWH7ZHewvcP
iOS: 100 msec – 200 msec (iPhone 7). https://drive.google.com/open?id=1lNhHue4xzQ7dHs24K_fSt84Os4bmNgZg
Android: 350 msec – 600 msec (Samsung Galaxy S8). https://drive.google.com/open?id=19uY2nq4rhTUiy9s2_DMUq8zV-29rpdtt
It used to be 100 msec – 200 msec in the previous Felgo version. Unfortunately, I do not have a screenshot of this.
Maybe this related to changes in Felgo or in the Qt version regarding Android?