How to convert mobilenet model into tflite? about aimark

[ibelem]

Hi @prajurock thanks for the question, you can get the tflite format models directly from https://github.com/ibelem/aimark/blob/master/static/model/README.md

What's your original model format of mobilenet?

[ibelem]

@prajurock BTW, i saw you are working on something with posenet, you can try: https://intel.github.io/webml-polyfill/examples/skeleton_detection/ This is the open source project that Intel is working on (https://github.com/intel/webml-polyfill)

TFLite mode for Posenet: https://www.tensorflow.org/lite/models/pose_estimation/overview

[prajurock]

yes I am working on deploy posenet tflite on mobile.

And I wanted to convert mobilenet 101 into tflite.

[ibelem]

@prajurock OK, seems like Google provided 075 rather than 101.

Usually please use TFLite convert cmdline and tflite_convert.py to convert models into tflite format. Do you have original 101 .pb file for posenet? Then you can convert the .pb file to .tflite.

If you have no any 101 .pb file, there may be another approach: Convert existed Tensorflow.js 101 model to tflite, maybe the posenet-python project can do this but we didn't try.

Posenet 101 tensorflow.js file
https://storage.googleapis.com/tfjs-models/weights/posenet/mobilenet_v1_101/manifest.json

https://storage.googleapis.com/tfjs-models/weights/posenet/mobilenet_v1_101/
├── manifest.json
├── MobilenetV1_Conv2d_0_biases
├── MobilenetV1_Conv2d_0_weights
├── MobilenetV1_Conv2d_10_depthwise_biases
├── MobilenetV1_Conv2d_10_depthwise_depthwise_weights
├── MobilenetV1_Conv2d_10_pointwise_biases
├── MobilenetV1_Conv2d_10_pointwise_weights
├── MobilenetV1_Conv2d_11_depthwise_biases
├── MobilenetV1_Conv2d_11_depthwise_depthwise_weights
├── MobilenetV1_Conv2d_11_pointwise_biases
├── MobilenetV1_Conv2d_11_pointwise_weights
├── MobilenetV1_Conv2d_12_depthwise_biases
├── MobilenetV1_Conv2d_12_depthwise_depthwise_weights
├── MobilenetV1_Conv2d_12_pointwise_biases
├── MobilenetV1_Conv2d_12_pointwise_weights
├── MobilenetV1_Conv2d_13_depthwise_biases
├── MobilenetV1_Conv2d_13_depthwise_depthwise_weights
├── MobilenetV1_Conv2d_13_pointwise_biases
├── MobilenetV1_Conv2d_13_pointwise_weights
├── MobilenetV1_Conv2d_1_depthwise_biases
├── MobilenetV1_Conv2d_1_depthwise_depthwise_weights
├── MobilenetV1_Conv2d_1_pointwise_biases
├── MobilenetV1_Conv2d_1_pointwise_weights
├── MobilenetV1_Conv2d_2_depthwise_biases
├── MobilenetV1_Conv2d_2_depthwise_depthwise_weights
├── MobilenetV1_Conv2d_2_pointwise_biases
├── MobilenetV1_Conv2d_2_pointwise_weights
├── MobilenetV1_Conv2d_3_depthwise_biases
├── MobilenetV1_Conv2d_3_depthwise_depthwise_weights
├── MobilenetV1_Conv2d_3_pointwise_biases
├── MobilenetV1_Conv2d_3_pointwise_weights
├── MobilenetV1_Conv2d_4_depthwise_biases
├── MobilenetV1_Conv2d_4_depthwise_depthwise_weights
├── MobilenetV1_Conv2d_4_pointwise_biases
├── MobilenetV1_Conv2d_4_pointwise_weights
├── MobilenetV1_Conv2d_5_depthwise_biases
├── MobilenetV1_Conv2d_5_depthwise_depthwise_weights
├── MobilenetV1_Conv2d_5_pointwise_biases
├── MobilenetV1_Conv2d_5_pointwise_weights
├── MobilenetV1_Conv2d_6_depthwise_biases
├── MobilenetV1_Conv2d_6_depthwise_depthwise_weights
├── MobilenetV1_Conv2d_6_pointwise_biases
├── MobilenetV1_Conv2d_6_pointwise_weights
├── MobilenetV1_Conv2d_7_depthwise_biases
├── MobilenetV1_Conv2d_7_depthwise_depthwise_weights
├── MobilenetV1_Conv2d_7_pointwise_biases
├── MobilenetV1_Conv2d_7_pointwise_weights
├── MobilenetV1_Conv2d_8_depthwise_biases
├── MobilenetV1_Conv2d_8_depthwise_depthwise_weights
├── MobilenetV1_Conv2d_8_pointwise_biases
├── MobilenetV1_Conv2d_8_pointwise_weights
├── MobilenetV1_Conv2d_9_depthwise_biases
├── MobilenetV1_Conv2d_9_depthwise_depthwise_weights
├── MobilenetV1_Conv2d_9_pointwise_biases
├── MobilenetV1_Conv2d_9_pointwise_weights
├── MobilenetV1_displacement_bwd_2_biases
├── MobilenetV1_displacement_bwd_2_weights
├── MobilenetV1_displacement_fwd_2_biases
├── MobilenetV1_displacement_fwd_2_weights
├── MobilenetV1_heatmap_2_biases
├── MobilenetV1_heatmap_2_weights
├── MobilenetV1_offset_2_biases
└── MobilenetV1_offset_2_weights

[prajurock]

@ibelem ThankYou very much