Mike discovered that an app uses an AlexNet TFLite model every few seconds to identify which game you are playing

Mike's inspection revealed that an app runs an INT8-quantized TensorFlow Lite model on 960×540 images every few seconds to determine which game the user is playing. The model is named alex.tflite and is in fact AlexNet, with a structure that appears to follow torchvision.models.alexnet. It performs nearest-neighbor feature matching against a database, but according to Mike's observations it fails to work in 99% of cases, and the results are discarded before the consent authorization step.