3 _tf2tfliteV2_ is a TensorFlow to TensorFlow Lite model Converter.
5 ## Where does V2 come from?
6 Even though we alreay have _tf2tflite_, we cannot cover all opeartors in TensorFlow. To expand coverage, we introduce _tf2tfliteV2_ which uses `TensorFlow Lite Converter`(by Google) internally.
9 - Frozen graph from TensorFlow 1.13.1 in binary(`*.pb`) or text(`*.pbtxt`) format
10 - Desired version of TensorFlow(You can use python virtualenv, docker, etc.)
14 python tf2tfliteV2.py \
16 > -i frozen_graph.pb -o converted.tflite
17 > -I model_inputs -O model_outputs
20 python tf2tfliteV2.py \
22 > --input_path=frozen_graph.pb \
23 > --output_path=converted.tflite \
24 > --input_arrays=model_inputs \
25 > --output_arrays=model_outputs
29 python tf2tfliteV2.py \
31 > --input_path=frozen_graph.pbtxt \
32 > --output_path=converted.tflite \
33 > --input_arrays=model_inputs \
34 > --output_arrays=model_outputs
37 python tf2tfliteV2.py \
39 > --input_path=multiple_output_graph.pb \
40 > --output_path=converted.tflite \
41 > --input_arrays=model_inputs \
42 > --output_arrays=output,output:1,output:2
47 -h, --help show this help message and exit
48 --v1 Use TensorFlow Lite Converter 1.x
49 --v2 Use TensorFlow Lite Converter 2.x
50 --graph_def Use graph def file(default)
51 --saved_model Use saved model
52 --keras_model Use keras model
53 -i INPUT_PATH, --input_path INPUT_PATH
54 Full filepath of the input file.
55 -o OUTPUT_PATH, --output_path OUTPUT_PATH
56 Full filepath of the output file.
57 -I INPUT_ARRAYS, --input_arrays INPUT_ARRAYS
58 Names of the input arrays, comma-separated.
59 -s INPUT_SHAPES, --input_shapes INPUT_SHAPES
60 Shapes corresponding to --input_arrays, colon-
61 separated.(ex:"1,4,4,3:1,20,20,3")
62 -O OUTPUT_ARRAYS, --output_arrays OUTPUT_ARRAYS
63 Names of the output arrays, comma-separated.