README.md

examples

English Version

Tengine Lite 的 examples 将提供简单的、好玩的 demo。

• 分类任务
• 人脸关键点检测任务
• ssd 目标检测任务
• retinaface 人脸检测任务
• scrfd 人脸检测任务
• yolact 实例分割任务
• unet 图像分割任务
• yolov3 目标检测任务
• yolov4-tiny 目标检测任务
• yolov5s 目标检测任务
• nanodet 目标检测任务
• efficientdet 目标检测任务
• yolox 目标检测任务
• openpose 人体姿态识别任务
• hrnet 人体姿态识别任务
• crnn 汉字识别任务
• seghuman 人像分割任务

除单张图片单模型推理级的任务外，Tengine Lite 还提供了基于视频流/图片流 pipeline 级别的功能演示

• 人体距离预测
• 人脸注册

---

分类任务 - tm_classification.c

Tengine Lite 兼容 Tengine 原有的 C API 供用户使用，这里我们使用 C API 展示如何运行 tm_classification 例程运行 MobileNet v1 分类网络模型，实现指定图片分类的功能。让你快速上手 Tengine Lite C API。这里，我们使用在这个撸猫时代行业从业者大爱的 tiger cat 作为测试图片。

模型仓库

模型仓库包含了运行examples所需模型、图片和文档。

• 百度网盘 （提取码：7ke5）

• Google Drive

源码参考

tm_classification.c

编译

build.sh 编译脚本默认配置已实现自动编译 examples 中的 demo 程序，以 x86 平台为例，demo 存放在 ./build/install/bin/ 目录下。

bug1989@DESKTOP-SGN0H2A:/mnt/d/ubuntu/gitlab/build-linux$ tree install
install
├── bin
│   ├── tm_alphapose
│   ├── tm_classification
│   ├── tm_classification_int8
│   ├── tm_classification_uint8
│   ├── tm_crnn
│   ├── tm_efficientdet
│   ├── tm_efficientdet_uint8
│   ├── tm_hrnet
│   ├── tm_landmark
│   ├── tm_landmark_uint8
│   ├── tm_mobilefacenet
│   ├── tm_mobilefacenet_uint8
│   ├── tm_mobilenet_ssd
│   ├── tm_mobilenet_ssd_uint8
│   ├── tm_nanodet_m
│   ├── tm_openpose
│   ├── tm_retinaface
│   ├── tm_scrfd
│   ├── tm_ultraface
│   ├── tm_unet
│   ├── tm_yolact
│   ├── tm_yolact_uint8
│   ├── tm_yolofastest
│   ├── tm_yolov3
│   ├── tm_yolov3_tiny
│   ├── tm_yolov3_tiny_uint8
│   ├── tm_yolov3_uint8
│   ├── tm_yolov4
│   ├── tm_yolov4_tiny
│   ├── tm_yolov4_tiny_uint8
│   ├── tm_yolov4_uint8
│   ├── tm_yolov5
│   └── tm_yolov5s
├── include
│   └── tengine
│       └── c_api.h
└── lib
    ├── libtengine-lite-static.a
    └── libtengine-lite.so

运行结果

将测试图片和模型文件放在 Tengine-Lite 根目录下，运行：

$ export LD_LIBRARY_PATH=./build/install/lib
$ ./build/install/bin/tm_classification -m models/mobilenet.tmfile -i images/cat.jpg -g 224,224 -s 0.017,0.017,0.017 -w 104.007,116.669,122.679

结果如下：

tengine-lite library version: 1.4-dev

model file : models/mobilenet.tmfile
image file : images/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 33.74 ms, max_time 33.74 ms, min_time 33.74 ms
--------------------------------------
8.574144, 282
7.880117, 277
7.812573, 278
7.286458, 263
6.357486, 281
--------------------------------------

人脸关键点检测任务 - tm_landmark.cpp

使用图片：

$ export LD_LIBRARY_PATH=./build/install/lib
$ ./build/install/bin/tm_landmark -m models/landmark.tmfile -i images/mobileface02.jpg -r 1 -t 1

结果如下：

tengine-lite library version: 1.4-dev
Repeat [1] min 8.784 ms, max 8.784 ms, avg 8.784 ms

ssd 目标检测任务 - tm_mobilenet_ssd.c

使用图片：

$ export LD_LIBRARY_PATH=./build/install/lib
$ ./build/install/bin/tm_mobilenet_ssd -m models/mobilenet_ssd.tmfile -i images/ssd_dog.jpg -r 1 -t 1

结果如下：

tengine-lite library version: 1.4-dev
Repeat 1 times, thread 1, avg time 78.89 ms, max_time 78.89 ms, min_time 78.89 ms
--------------------------------------
detect result num: 3 
dog     :99.8%
BOX:( 138 , 209 ),( 324 , 541 )
car     :99.7%
BOX:( 467 , 72 ),( 687 , 171 )
bicycle :99.5%
BOX:( 107 , 141 ),( 574 , 415 )
======================================
[DETECTED IMAGE SAVED]:
======================================

retinaface 人脸检测任务 - tm_retinaface.cpp

使用图片：

$ export LD_LIBRARY_PATH=./build/install/lib
$ ./build/install/bin/tm_retinaface -m models/retinaface.tmfile -i images/mtcnn_face4.jpg -r 1 -t 1

结果如下：

tengine-lite library version: 1.4-dev
img_h, img_w : 316, 474
Repeat 1 times, thread 1, avg time 28.78 ms, max_time 28.78 ms, min_time 28.78 ms
--------------------------------------
detected face num: 4
BOX 1.00:( 38.4053 , 86.142 ),( 46.3009 , 64.0174 )
BOX 0.99:( 384.076 , 56.9844 ),( 76.968 , 83.9609 )
BOX 0.99:( 169.196 , 87.1324 ),( 38.4133 , 46.8504 )
BOX 0.98:( 290.004 , 104.453 ),( 37.6346 , 46.7777 )

scrfd 人脸检测任务 - tm_scrfd.cpp

使用图片：

$ export LD_LIBRARY_PATH=./build/install/lib
$ ./build/install/bin/tm_scrfd -m models/scrfd_2.5g_kps.tmfile -i images/face5.jpg -r 1 -t 1

结果如下：

tengine-lite library version: 1.5-dev
Repeat 1 times, thread 1, avg time 289.97 ms, max_time 289.97 ms, min_time 289.97 ms
--------------------------------------
detection num: 5
0.90917 at 199.37 54.92 28.52 x 38.12
0.89985 at 70.50 29.96 32.26 x 41.25
0.88838 at 111.36 48.00 33.53 x 46.77
0.88484 at 247.54 51.15 30.21 x 37.29
0.83953 at 149.23 49.48 27.89 x 38.50

yolact 实例分割任务 - tm_yolact.cpp

使用图片：

$ export LD_LIBRARY_PATH=./build/install/lib
$ ./build/install/bin/tm_yolact -m models/yolact.tmfile -i images/ssd_car.jpg -r 1 -t 1

结果如下：

tengine-lite library version: 1.4-dev
Repeat 1 times, thread 1, avg time 2064.44 ms, max_time 2064.44 ms, min_time 2064.44 ms
--------------------------------------
6 = 0.99966 at 130.82 57.77 340.78 x 237.36
3 = 0.99675 at 323.39 194.97 175.57 x 132.96
1 = 0.33431 at 191.24 195.78 103.06 x 179.22

unet 图像分割任务 - tm_unet.cpp

使用图片：

$ export LD_LIBRARY_PATH=./build/install/lib
$ ./build/install/bin/tm_unet -m models/unet_sim3.tmfile -i images/carvana01.jpg -r 1 -t 1

结果如下：

Image height not specified, use default 512
Image width not specified, use default  512
Scale value not specified, use default  0.00392, 0.00392, 0.00392
tengine-lite library version: 1.4-dev

model file : models/unet_sim3.tmfile
image file : images/carvana01.jpg
img_h, img_w, scale[3], mean[3] : 512 512 , 0.004 0.004 0.004, 0.0 0.0 0.0
Repeat 1 times, thread 1, avg time 4861.93 ms, max_time 4861.93 ms, min_time 4861.93 ms
--------------------------------------
segmentation result is save as unet_out.png

yolov3 目标检测任务 - tm_yolov3.cpp

使用图片：

$ export LD_LIBRARY_PATH=./build/install/lib
$ ./build/install/bin/tm_yolov3 -m models/yolov3.tmfile -i images/ssd_dog.jpg -r 1 -t 1

结果如下：

tengine-lite library version: 1.4-dev
Repeat 1 times, thread 1, avg time 1131.67 ms, max_time 1131.67 ms, min_time 1131.67 ms
--------------------------------------
detection num: 3
16: 100%, [ 123,  223,  320,  544], dog
 1:  99%, [ 160,  117,  568,  435], bicycle
 7:  94%, [ 473,   87,  693,  166], truck

yolov4-tiny目标检测任务 - tm_yolov4_tiny.cpp

使用图片：

$ export LD_LIBRARY_PATH=./build/install/lib
$ ./build/install/bin/tm_yolov4_tiny -m models/yolov4-tiny.tmfile -i images/ssd_dog.jpg -r 1 -t 1

结果如下：

tengine-lite library version: 1.4-dev
Repeat 1 times, thread 1, avg time 152.50 ms, max_time 152.50 ms, min_time 152.50 ms
--------------------------------------
detection num: 3
16:  87%, [ 136,  206,  318,  542], dog
 7:  81%, [ 463,   79,  703,  170], truck
 1:  61%, [  72,  100,  577,  479], bicycle

yolov5s目标检测任务 - tm_yolov5s.cpp

使用图片：

$ export LD_LIBRARY_PATH=./build/install/lib
$ ./build/install/bin/tm_yolov5s -m models/yolov5s.tmfile -i images/ssd_dog.jpg -r 1 -t 1

结果如下：

tengine-lite library version: 1.4-dev
Repeat 1 times, thread 1, avg time 462.94 ms, max_time 462.94 ms, min_time 462.94 ms
--------------------------------------
detection num: 3
16:  89%, [ 135,  218,  313,  558], dog
 7:  86%, [ 472,   78,  689,  169], truck
 1:  75%, [ 123,  107,  578,  449], bicycle

nanodet目标检测任务 - tm_nanodet_m.cpp

使用图片：

$ export LD_LIBRARY_PATH=./build/install/lib
$ ./build/install/bin/tm_nanodet_m -m models/nanodet.tmfile -i images/ssd_dog.jpg -r 1 -t 1

结果如下：

tengine-lite library version: 1.4-dev
Repeat 1 times, thread 1, avg time 35.96 ms, max_time 35.96 ms, min_time 35.96 ms
--------------------------------------
detection num: 3
 1: 59.313%, [141.945, 160.890, 563.568, 429.829], bicycle
16: 50.605%, [132.646, 205.861, 312.255, 511.470], dog
 2: 48.931%, [462.477,  72.462, 701.777, 170.343], car

efficientdet目标检测任务 - tm_efficientdet.c

使用图片：

$ export LD_LIBRARY_PATH=./build/install/lib
$ ./build/install/bin/tm_efficientdet -m ../models/efficientdet.tmfile -i ../images/ssd_dog.jpg -r 1 -t 1

结果如下：

tengine-lite library version: 1.4-dev
model file : ../models/efficientdet.tmfile
image file : ../images/ssd_dog.jpg
img_h, img_w, scale[3], mean[3] : 512 512 , 0.017 0.018 0.017, 123.7 116.3 103.5
Repeat 1 times, thread 1, avg time 598.86 ms, max_time 598.86 ms, min_time 598.86 ms
--------------------------------------
17:  80%, [ 132,  222,  315,  535], dog
 7:  73%, [ 467,   74,  694,  169], truck
 1:  42%, [ 103,  119,  555,  380], bicycle
 2:  29%, [ 687,  113,  724,  156], car
 2:  25%, [  57,   77,  111,  124], car

yolox目标检测任务 - tm_yolox.cpp

使用图片：

$ export LD_LIBRARY_PATH=./build/install/lib
$ ./build/install/bin/tm_yolox -m ../models/yolox_nano.tmfile -i ../images/ssd_dog.jpg -r 1 -t 1

结果如下：

tengine-lite library version: 1.5-dev
Repeat 1 times, thread 1, avg time 97.64 ms, max_time 97.64 ms, min_time 97.64 ms
--------------------------------------
detection num: 3
16:  85%, [ 132,  216,  318,  545], dog
 1:  83%, [ 112,  140,  568,  427], bicycle
 2:  69%, [ 466,   77,  693,  168], car

openpose人体姿态识别任务 - tm_openpose.cpp

使用图片：

$ export LD_LIBRARY_PATH=./build/install/lib
$ ./build/install/bin/tm_openpose -m models/openpose_coco.tmfile -i images/pose.jpg -r 1 -t 1

结果如下：

tengine-lite library version: 1.4-dev
Repeat 1 times, thread 1, avg time 7296.71 ms, max_time 7296.71 ms, min_time 7296.71 ms
--------------------------------------
KeyPoints Coordinate:
0:[292.174, 55.6522]
1:[306.087, 125.217]
2:[250.435, 139.13]
3:[236.522, 222.609]
4:[222.609, 306.087]
5:[361.739, 125.217]
6:[403.478, 208.696]
7:[417.391, 292.174]
8:[264.348, 306.087]
9:[264.348, 431.304]
10:[264.348, 570.435]
11:[347.826, 306.087]
12:[375.652, 431.304]
13:[333.913, 542.609]
14:[278.261, 41.7391]
15:[306.087, 41.7391]
16:[264.348, 55.6522]
17:[320, 55.6522]

人体姿态识别结果会保存为图片，名称为：Output-Keypionts.jpg和Output-Skeleton.jpg。

hrnet人体姿态识别任务 - tm_hrnet.cpp

使用图片：

$ export LD_LIBRARY_PATH=./build/install/lib
$ ./build/install/bin/tm_hrnet -m models/hrnet.tmfile -i images/pose.jpg -r 1 -t 1

结果如下：

tengine-lite library version: 1.4-dev
Repeat [1] min 416.223 ms, max 416.223 ms, avg 416.223 ms
x: 27, y: 58, score: 0.91551
x: 27, y: 45, score: 0.865156
x: 28, y: 30, score: 0.831916
x: 34, y: 29, score: 0.839507
x: 38, y: 44, score: 0.88559
x: 35, y: 55, score: 0.891349
x: 31, y: 30, score: 0.873104
x: 31, y: 14, score: 0.928233
x: 30, y: 10, score: 0.948434
x: 29, y: 1, score: 0.915752
x: 23, y: 31, score: 0.811694
x: 24, y: 24, score: 0.935574
x: 24, y: 14, score: 0.899991
x: 37, y: 13, score: 0.908696
x: 41, y: 22, score: 0.902927
x: 41, y: 29, score: 0.847032

汉字识别任务 - tm_crnn.cpp

模型文件：crnn_lite_dense.tmfile 测试图片：o2_resize.jpg 字库文件：keys.txt 测试图片：

$ export LD_LIBRARY_PATH=./build/install/lib
$ ./build/install/bin/tm_crnn -m models/crnn_lite_dense.tmfile -i images/o2_resize.jpg -l files/keys.txt

结果如下：

tengine-lite library version: 1.4-dev
Repeat 1 times, thread 1, avg time 23.30 ms, max_time 23.30 ms, min_time 23.30 ms
--------------------------------------
如何突破自己的颜值上限
--------------------------------------

其中ocr的识别结果会直接打印到终端中, 同时如果需要保存为txt文件可以修改源码使其重定向到文件。

人像分割任务 - tm_seghuman.cpp

模型文件：paddleSegSim.tmfile

测试图片：human_image.jpg

测试图片：

$ export LD_LIBRARY_PATH=./build/install/lib
$ ./build/install/bin/tm_seghuman -m models/paddleSegSim.tmfile -i images/human_image.jpeg

结果如下：

tengine-lite library version: 1.5-dev
Repeat 1 times, avg time 123.766 ms, max_time 123.766 ms, min_time 123.766 ms

人像分割结果会保存为图片，名称为：seg_human_result.jpg

人体距离预测

模型文件：mobilenet_ssd.tmfile

执行（建议用 GPU 推理）：

$ cd build/examples
$ ln -s models/mobilenet_ssd.tmfile
$ export LD_LIBRARY_PATH=./build/install/lib
$ ./tm_pipeline_estimate_ped_distance
detect result num: 1 
person	:100.0%
BOX:( 35 , 78 ),( 587 , 478 )
...

人脸注册

模型文件列表：

• rfb-320.tmfile 人脸检测
• landmark.tmfile 关键点
• mobilefacenet.tmfile 特征提取

$ cd build/examples
$ ln -s models/rfb-320.tmfile
$ ln -s models/landmark.tmfile
$ ln -s models/mobilefacenet.tmfile

假设注册集在 build/examples/images。 执行（建议用 GPU 推理）：

$ export LD_LIBRARY_PATH=./build/install/lib
$ ./tm_pipeline_enroll_face  ./images

当前目录会生成类似feature0.bin 多个序列化文件，存有人脸特征。

我们将持续更新各种有趣的 demo ，敬请期待......