Tengine Lite Quick Start Guide
Tengine Lite is an excellent lightweight end-to-side/embedded environment deep neural network inference engine. Compatible with multiple operating systems and deep learning algorithms, AIoT development kit based on AI reasoning framework. This document will take the classification model (TensorFlow MobileNetv1 model) as an example on the x86 Linux platform to take you quickly to get started with Tengine Lite.
$ git clone https://github.com/OAID/Tengine/ Tengine-Lite
Before starting to compile Tengine-Lite, you need to confirm that you have installed cmake, g++, if not, you can install it through the following instructions:
$ sudo apt install cmake g++-
We provide tmfile files of common open source models, you can find them here Tengine Model zoo(Password: hhgc )
-
You can also use convert_model_to_tm to convert the mobilenet.pb model into a Tengine model. Refer to Model Conversion Tool
When you get the Tengine model file such as mobilenet.tmfile, you can use Tengine Lite to happily develop applications on various platforms.
The core API of Tengine Lite is as follows:
- init_tengine
Initialize Tengine Lite, this function only needs to be called once in the program.
- create_graph
Create Tengine Lite calculation graph.
- prerun_graph
Pre-run, prepare the resources required for computational graph inference.
- run_graph
Start Tengine Lite calculation graph inference.
- postrun_graph
Stop the graph and release the resources occupied by the graph.
- destroy_graph
Destroy the graph.
The postrun_graph and destroy_graph are called after the model inference is executed, and they are generally called continuously.
Tengine provides C and C++ API for users to use. Here we use Tengine C++ API to show how to run the MobileNetv1 network model to realize the image classification function, so that you can quickly get started with Tengine C++ API. We will use detailed code comments in the code to facilitate you to familiarize yourself with the functions of Tengine's core API and develop your own code faster. Here, we use the tiger cat picture that is loved by industry practitioners in this cat era as the test picture.
Example of test code: ./example/tm_classificaton.c
Refer to build.sh in the root directory, take local compilation (x86) as an example, enter the command as follows:
mkdir -p build
pushd build
cmake ..
make -j4 && make install
popdAfter the compilation is complete, if there is a libtengine-lite.so file in the build/install/lib directory, the compilation is successful.
ls /build/install/lib
libtengine-lite.soThe classification demo (tm_classification) now exists in the ./build/install/bin directory.
Find mobilenet.tmfile and synset_words.txt in the folder Tengine_models/classification/mobilenet/ you just downloaded.
Download the test picture (cat.jpg) and the model file (mobilenet.tmfile) in the root directory of Tengine-Lite,
Just run the following command:
export LD_LIBRARY_PATH=./build/install/lib
./build/install/example/tm_classification -m mobilenet.tmfile -i cat.jpg -g 224,224 -s 0.017,0.017,0.017 -w 104.007,116.669,122.679The results are as follows:
tengine-lite library version: 1.4-dev
model file: mobilenet.tmfile
image file: cat.jpg
img_h, img_w, scale[3], mean[3]: 224 224, 0.017 0.017 0.017, 104.0 116.7 122.7
Repeat 1 times, thread 1, avg time 30.53 ms, max_time 30.53 ms, min_time 30.53 ms
--------------------------------------
8.574144, 282
7.880117, 277
7.812573, 278
7.286458, 263
6.357486, 281
--------------------------------------Combined with the synset_words.txt file, we have successfully classified the test picture as a tiger cat. At this point, the most basic started guide has been completed. Friends can explore on their own examples. We will also update various tutorial examples from time to time to everyone~
...\(^0^)/...