[PYTHON] Judge Yosakoi Naruko by image classification of Tensorflow.

Note

I tried to classify by tensorflow, so I will leave a note. When classifying images in the environment of a fairly old PC (Ubuntu is installed on a PC about 10 years ago) It seems that it will not work according to the CPU specifications unless the tensor flow version is set to 1.5.

• Sites referred to for image classification https://blog.ch3cooh.jp/entry/20180930/1538287200

Rough explanation of file processing contents

1. Create a learning model by classifying with a file called retrain.py. This time, I learned the image of Naruko and the image of a cat. (The images of cats and Naruko are put in folders)

2. Judge the test image by referring to the training model in a file called label_image.py. This time, I prepared Naruko for the test image and judged it.

In addition, the location such as hierarchy is passed as a parameter to each file.

pip install tensorflow==1.5

pip install tensorflow_hub==0.2.0

python retrain.py \
  --bottleneck_dir=bottlenecks \
  --how_many_training_steps=1000 \
  --model_dir=inception \
  --summaries_dir=training_summaries/basic \
  --output_graph=retrained_graph.pb \
  --output_labels=retrained_labels.txt \
  --image_dir=img

python label_image.py  \
  --graph=retrained_graph.pb  \
  --labels=retrained_labels.txt  \
  --output_layer=final_result  \
  --image=test_images/x.jpg  \
  --input_layer=Mul

Download the source from below https://github.com/tensorflow/tensorflow/tree/r1.5/tensorflow

wget https://raw.githubusercontent.com/tensorflow/tensorflow/r1.5/tensorflow/examples/label_image/label_image.py
wget https://raw.githubusercontent.com/tensorflow/tensorflow/r1.5/tensorflow/examples/image_retraining/retrain.py