TFProfiler

TFProfiler is an app that aims to profile TensorFlow Lite model or OpenCV DNN supported ones (*.onnx, *.caffemodel, *.pb, *.weights, *.t7, *.net, etc). Measure model performance with FPS, model initialization time, inference time, memory consumption, etc. You can tweak model inferences on Android smartphone with different accelerators:

•  CPU

•  GPU

•  NNAPI

•  HEXAGON

•  XNNPACK

The app has a built-in subset of public available image dataset Caltech 101 (http://www.vision.caltech.edu/Image_Datasets/Caltech101/). It may be used for testing model inferences.

Source code of the app can be found here: https://github.com/iglaweb/TFProfiler

The app is intended only for testing and conducting experiments on your Android smartphone.