ibelem's picture
Init Commit
f543607
`.npy` files under this folder are downloaded from TFLite model of DeepLab V3 MoblieNet V2 generated by following commands, which is licensed under the Apache 2.0.
### How to Generate DeepLab V3 MobileNet V2 TFLite model
Download and extract the origin [DeepLab V3 MobileNet V2 model](http://download.tensorflow.org/models/deeplabv3_mnv2_pascal_trainval_2018_01_29.tar.gz)
use following tools and commands to convert the frozen grahp to TFLite model.
Use tensorflow's [`optimize_for_inference`](https://github.com/tensorflow/tensorflow/blob/master/tensorflow/python/tools/optimize_for_inference.py) tool to generate a stripped frozen graph:
```
python3 -m tensorflow.python.tools.optimize_for_inference \
--input=./frozen_inference_graph.pb \
--output=./frozen_inference_graph_stripped.pb \
--frozen_graph=True \
--input_names="sub_7" \
--output_names="ArgMax"
```
Use ['TensorFlow Lite converter'](https://www.tensorflow.org/lite/convert) tool to convert stripped frozen graph to TFLite model:
```
tflite_convert \
--graph_def_file=frozen_inference_graph_stripped.pb \
--output_file=deeplab_mobilenetv2.tflite \
--output_format=TFLITE \
--input_format=TENSORFLOW_GRAPHDEF \
--input_arrays=sub_7 \
--output_arrays=ArgMax
```