Process a video stream on a mobile phone to detect objects in a museum.
Identification is possible through a lightweight neural network. The model should offer sufficient accuracy and speed in recognizing different types of exhibits (size, material) in diverse conditions (lighting, location, background, viewing angles). At the same time, it should consider the limitations of the mobile device, particularly the limited computing power, memory, and battery capacity.
Illustration taken from
https://viso.ai/wp-content/uploads/2022/06/mediapipe-object_tracking_android_gpu.gif
https://i.giphy.com/uULru6cnBO4gM.webp