[AI] Python3 uses TF-Hub for transfer learning (image classification)

machine environment

• win10
• python3.6

---

Install dependencies

• pip install tensorflow==1.7.0
• pip install tensorflow-hub

---

Github address

• https://github.com/ns2250225/tensorflow-hub-image-classify

---

Prepare picture material

• Note: Each category must have at least 100 pictures, otherwise an error will be reported
• Put the pictures into categories and put them in the photos folder

---

Train the model (default is to use Inception V3)

• Clear the output folder first
• 20 steps of training here

python .\retrain.py --image_dir .\photos\ --saved_model_dir .\mod
el\ --bottleneck_dir .\bottleneck\ --how_many_training_steps 20 --output_labels .\output\output_labels.txt --output_grap
h .\output\retrain.pb

---

Test the model (test a single image)

python .\label_image.py --graph .\output\retrain.pb --labels .\o
utput\output_labels.txt --input_layer=Placeholder --output_layer=final_result --image .\test.jpg

---

Remark

• Can use other models to train
• Add when training the model: --tfhub_module https://tfhub.dev/google/imagenet/mobilenet_v2_100_224/feature_vector/1