一、环境搭建
参考:官网入门
注意:软件安装过程中环境变量的配置。
二、模型测试(基于Ubuntu)
I.分类模型(以MobileNet V2 (ImageNet)为例说明)
- 板子供电,Type-C数据线与PC相连
- 下载Edge TPU模型文件(.tflite)和Labels文件(.txt)
- 将模型文件上传至板子,使用如下mdt命令把数据上传到设备model_test目录
mdt push dog.jpg model_test
mdt push imagenet_labels.txt model_test
mdt push mobilenet_v2_1.0_224_quant_edgetpu.tflite model_test
- 另开一个终端,通过mdt连接开发板
mdt devices //查看当前设备连接情况
mdt shell //进入连接设备控制台
- 在mdt控制台查看模型文件,已上传至开发板
[email protected]:~$ cd model_test/
[email protected]:~/model_test$ ls
dog.jpg imagenet_labels.txt mobilenet_v2_1.0_224_quant_edgetpu.tflite
- mdt控制台进入到有.py文件的demo目录,运行模型
[email protected]:~/model_test$ cd /usr/lib/python3/dist-packages/edgetpu/demo
[email protected]:/usr/lib/python3/dist-packages/edgetpu/demo$python3 classify_image.py \
--model ~/model_test/mobilenet_v2_1.0_224_quant_edgetpu.tflite \
--label ~/model_test/imagenet_labels.txt \
--image ~/model_test/dog.jpg
---------------------------
Border collie
Score : 0.964844
其中:dog.jpg 如下图
![](https://img.laitimes.com/img/9ZDMuAjOiMmIsIjOiQnIsIyZuBnLzADO5QTOyUTM3EzNwkTMwIzLc52YucWbp5GZzNmLn9Gbi1yZtl2Lc9CX6MHc0RHaiojIsJye.png)
II.目标检测模型(以MobileNet SSD v2 (Faces)为例说明)
- 前四步和Ⅰ中类似,直接到第五步;
- 在mdt控制台查看模型文件,已上传至开发板
[email protected]:~$ cd face_detect
[email protected]:~/face_detect$ ls
mobilenet_ssd_v2_face_quant_postprocess.tflite people.jpg
- mdt控制台运行模型
[[email protected] face_detect]# mdt pull people_result.jpg .
Waiting for a device...
Connecting to coy-umpire at 192.168.101.115
100% |<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<| people_result.jpg
[[email protected] face_detect]# ls
mobilenet_ssd_v2_face_quant_postprocess.tflite people.jpg people_result.jpg
- 本机控制台从板子上下载并查看结果
[email protected]:~/face_detect$ cd /usr/lib/python3/dist-packages/edgetpu/demo
[email protected]:/usr/lib/python3/dist-packages/edgetpu/demo$ python3 object_detection.py \
> --model ~/face_detect/mobilenet_ssd_v2_face_quant_postprocess.tflite \
> --input ~/face_detect/people.jpg \
> --output ~/people_result.jpg
-----------------------------------------
score = 0.996094
box = [34.58340883255005, 3.6951268389821053, 79.72809076309204, 54.911569476127625]
-----------------------------------------
score = 0.917969
box = [70.44789791107178, 0.24352699518203735, 116.58040523529053, 51.459967851638794]
-----------------------------------------
score = 0.691406
box = [0.6054049730300903, 7.107417978346348, 40.25177478790283, 62.95591643452644]
-----------------------------------------
score = 0.308594
box = [12.797946631908417, 93.51808422803879, 41.02548837661743, 126.22678685188293]
-----------------------------------------
score = 0.167969
box = [68.79348278045654, 71.8349220752716, 94.12553787231445, 100.55993902683258]
Please check /home/mendel/people_result.jpg
people.jpg 如下:
people_result.jpg如下:
模型来自:谷歌官网模型