Edited at

raspberry pi 1でtensorflow lite


概要

raspberry pi 1でtensorflow liteやってみた。

raspberry pi 1 model bに、インストールしてみた。


環境

Raspberry pi 1 model b(element14)

sd-card 8GB

raspbian 2016-11-25 jessie with pixel


インストール

git cloneして、armv6lでビルドした。


確認

binに、benchmark_modelとminimalができた。

libに、libtensorflow-lite.aができた。

./minimal xor_model.tflite

=== Pre-invoke Interpreter State ===

Interpreter has 13 tensors and 4 nodes
Inputs: 8
Outputs: 1

Tensor 0 activation/Tanh kTfLiteFloat32 kTfLiteArenaRw 32 bytes ( 0.0 MB) 1 8
Tensor 1 activation_1/Sigmoid kTfLiteFloat32 kTfLiteArenaRw 4 bytes ( 0.0 MB) 1 1
Tensor 2 dense/BiasAdd kTfLiteFloat32 kTfLiteArenaRw 32 bytes ( 0.0 MB) 1 8
Tensor 3 dense/MatMul_bias kTfLiteFloat32 kTfLiteMmapRo 32 bytes ( 0.0 MB) 8
Tensor 4 dense/kernel/transpose kTfLiteFloat32 kTfLiteMmapRo 64 bytes ( 0.0 MB) 8 2
Tensor 5 dense_1/BiasAdd kTfLiteFloat32 kTfLiteArenaRw 4 bytes ( 0.0 MB) 1 1
Tensor 6 dense_1/MatMul_bias kTfLiteFloat32 kTfLiteMmapRo 4 bytes ( 0.0 MB) 1
Tensor 7 dense_1/kernel/transpose kTfLiteFloat32 kTfLiteMmapRo 32 bytes ( 0.0 MB) 1 8
Tensor 8 dense_input kTfLiteFloat32 kTfLiteArenaRw 8 bytes ( 0.0 MB) 1 2
Tensor 9 (null) kTfLiteNoType kTfLiteMemNone 0 bytes ( 0.0 MB) (null)
Tensor 10 (null) kTfLiteNoType kTfLiteMemNone 0 bytes ( 0.0 MB) (null)
Tensor 11 (null) kTfLiteNoType kTfLiteMemNone 0 bytes ( 0.0 MB) (null)
Tensor 12 (null) kTfLiteNoType kTfLiteMemNone 0 bytes ( 0.0 MB) (null)

Node 0 Operator Builtin Code 9
Inputs: 8 4 3
Outputs: 2
Node 1 Operator Builtin Code 28
Inputs: 2
Outputs: 0
Node 2 Operator Builtin Code 9
Inputs: 0 7 6
Outputs: 5
Node 3 Operator Builtin Code 14
Inputs: 5
Outputs: 1

./benchmark_model --graph=xor_model.tflite

STARTING!

Min num runs: [50]
Min runs duration (seconds): [1]
Inter-run delay (seconds): [-1]
Num threads: [1]
Benchmark name: []
Output prefix: []
Min warmup runs: [1]
Min warmup runs duration (seconds): [0.5]
Graph: [xor_model.tflite]
Input layers: []
Input shapes: []
Use nnapi : [0]
Loaded model xor_model.tflite
resolved reporter
Initialized session in 4.23ms
Running benchmark for at least 1 iterations and at least 0.5 seconds
count=2963 first=385 curr=19 min=18 max=20218 avg=30.4077 std=377

Running benchmark for at least 50 iterations and at least 1 seconds
count=7331 first=43 curr=20 min=17 max=298 avg=22.4294 std=7

Average inference timings in us: Warmup: 30.4077, Init: 4230, no stats: 22.4294

以上。