Tensorflow之图像识别

1、Tensorflow 环境搭建
为了简单，直接使用docker

# docker pull tensorflow/tensorflow:latest
# sudo mkdir -p /data/tensorflow/notebooks
# docker run -it --name myts -v /data/tensorflow/notebooks:/notebooks -p 8888:8888 daocloud.io/daocloud/tensorflow:latest

进入容器

# sudo docker exec -it d708c1700244 /bin/bash
# cd /home/ 
# mkdir lyj

2、下载 TensorFlow 模型代码库
在宿主机中下载，再拷到docker容器中，其实在容器中下载也是一样的

# git clone https://github.com/tensorflow/models.git
# docker cp models d708c1700244:/home/lyj

3、在容器中运行

# cd models/tutorials/image/imagenet
# python classify_image.py

识别结果

giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca (score = 0.89107)
indri, indris, Indri indri, Indri brevicaudatus (score = 0.00779)
lesser panda, red panda, panda, bear cat, cat bear, Ailurus fulgens (score = 0.00296)
custard apple (score = 0.00147)
earthstar (score = 0.00117)

4、识别自定义图片
从网上下载一张图片，命名为 5.jpg

拷贝到容器中

# docker cp 5.jpg d708c1700244:/home/lyj/models/tutorials/image/imagenet/

修改代码默认图片

# vim classify_image.py

parser.add_argument(
  '--image_file',
  type=str,
  default='./5.jpg',
  help='Absolute path to image file.'
  )

识别结果

# python classify_image.py

airliner (score = 0.91857)
wing (score = 0.03277)
warplane, military plane (score = 0.00107)
space shuttle (score = 0.00062)
viaduct (score = 0.00048)

识别出 airliner(客机)，得分0.91857，还是挺准的

参考：
https://blog.csdn.net/freewebsys/article/details/70237003
https://tensorflow.google.cn/tutorials/images/image_recognition