paddle lite 在 x86 平台运行的python示例

本文是通过稍微改造官方 Paddle-Lite-Demo 库里的例子来运行的

1. 下载预测库

python -m pip install paddlelite

2. 下载 Paddle-Lite-Demo

git clone https://github.com/PaddlePaddle/Paddle-Lite-Demo.git

cd Paddle-Lite-Demo
mkdir _data     # 创建 _data 存放模型文件和转换后的模型文件
cd _data

3. 模型下载

wget http://paddle-inference-dist.bj.bcebos.com/mobilenet_v1.tar.gz
tar zxf mobilenet_v1.tar.gz

4. 模型转换

paddle_lite_opt --model_dir=./mobilenet_v1 --optimize_out_type=naive_buffer --optimize_out=./mobilenet_v1_opt --valid_targets=x86

paddle_lite_opt是安装上面安装paddlelite后就有的工具

注意--valid_targets=x86，选择x86平台

此时会生成转换后的文件 mobilenet_v1_opt.nb

5. 改变demo里的参数来运行图像分类任务

进入相关目录，运行如下命令

cd ..
cd image_classification/armlinux/shell/python

python ./image_classification.py --model_dir ../../../../_data/mobilenet_v1_opt.nb \
  --input_shape=1,3,224,224 --image_path ../../../assets/images/tabby_cat.jpg \
  --label_path ../../../assets/labels/labels.txt \
  --topk=3 --repeats=100 --warmup=10

如果有如下报错，则把image_classification.py文件的79行config.set_threads(threads)注释掉，改为# config.set_threads(threads)

Traceback (most recent call last):
  File "./image_classification.py", line 197, in <module>
    RunModel(args)
  File "./image_classification.py", line 156, in RunModel
    predictor = Init(model_dir, thread_num)
  File "./image_classification.py", line 79, in Init
    config.set_threads(threads)
AttributeError: 'paddlelite.lite.MobileConfig' object has no attribute 'set_threads'

把--model_dir设置为上一步生成的nb文件路径../../../../_data/mobilenet_v1_opt.nb

会得到分类结果如下：

    ================== Speed Report ===================
    model: ../../../../_data/mobilenet_v1_opt.nb, run avg_time:  1.2612977027893067e-05 ms, min_time:  1.2119770050048829e-05 ms
    ================== Precision Report ===================
    i: 0, index: 285, name: cat
    , score: 0.4729938805103302
    i: 1, index: 281, name: cat
    , score: 0.4363744556903839
    i: 2, index: 282, name: cat
    , score: 0.0766439363360405
    ================== Report End ===================