How to set xnnpack num_threads in mediapipe python?

[mch0dmin]

In mediapipe C++, i can set xnnpack num_threads in .pbtxt. For example,

136 # Runs model inference on CPU. 137 node { 138 calculator: "InferenceCalculator" 139 input_side_packet: "MODEL:model" 140 input_stream: "TENSORS:input_tensors" 141 output_stream: "TENSORS:output_tensors" 142 options: { 143 [mediapipe.InferenceCalculatorOptions.ext] { 144 delegate { xnnpack { num_threads: 32 }} 145 #delegate { xnnpack {} } 146 } 147 } 148 } But how to set it in mediapipe python?

Hi @sureshdagooglecom, can you give me some advices? Thanks!

[mch0dmin]

hi @kuaashish, @sureshdagooglecom , Can you give me some advices?

[postlund]

I'm interested in this as well. Doesn't seem possible to modify with current API?

[ayushgdev]

Hello @mch0dmin You can pass it in the calculator params of InferenceCalculator using Python, however, it is not available to simply "pass" the value. You need to make several changes:

1. Depending on the solution you are using, modify the __init__() of that solution to accept a new parameter for the xnnpack num_threads value. For example, if using 'FaceMesh' you need to add an argument here
2. Change the calculator params argument to pass on the new xnnpack num_threads value. It should be something like this (taking FaceMesh example):

calculator_params={ 
'facedetectionshortrangecpu__facedetectionshortrange__facedetection__TensorsToDetectionsCalculator.min_score_thresh': min_detection_confidence,
 'facelandmarkcpu__ThresholdingCalculator.threshold':min_tracking_confidence,
'facedetectionshortrangecpu__facedetectionshortrange__facedetection__InferenceCalculator.xnnpack': {'num_threads': num_threads}
},

3. Add option_value: xnnpack:options/xnnpack to the InferenceCalculator in face_detection.pbtxt file
4. Rebuild the Python package by following the instructions given here

[google-ml-butler[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you.

[google-ml-butler[bot]]

Closing as stale. Please reopen if you'd like to work on this further.

[google-ml-butler[bot]]

Are you satisfied with the resolution of your issue? Yes No