tensorflow-image-classifier

Use Tensorflow to train and learn images

The original implementation can be found at https://codelabs.developers.google.com/codelabs/tensorflow-for-poets/#0

We'll be using mobilenet with parameters 0.50 and 224 in this example.

Folders layout:

• photos: contains all the photos for training the model
• scripts: will be calling the files here to run our model
• tf_files: outputs from the model
• uploaded_photos: dump photos that we want to classify into this folder.

Opening Tensorboard

tensorboard --logdir=tf_files/training_summaries &

Training images

Run this to start training images:

python -m scripts.retrain --bottleneck_dir=tf_files/bottlenecks --model_dir=tf_files/models/ --summaries=tf_files/training_summaries/mobilenet_0.50_224 --output_graph=tf_files/retrained_graph.pb --output_labels=tf_files/retrained_labels.txt --architecture=mobilenet_0.50_224 --image_dir=photos

This should take a little while.

Classifying images

Let's classify an iPhone 7 plus.

Let's use an annoymous photo of an iPhone 7 plus to validate our trained model:

python -m scripts.label_image --graph=tf_files/retrained_graph.pb --image=uploaded_photos/photo1.jpg

You should get an output something like this:

iphone 7 plus 0.989413
xiaomi redmi 4a 0.00497179
tulips 0.00374058
roses 0.00186522
sunflowers 5.5651e-06

Let's classify a Xiaomi Redmi 4A

python -m scripts.label_image --graph=tf_files/retrained_graph.pb --image=uploaded_photos/photo2.jpg

You should get an output something like this:

xiaomi redmi 4a 0.999836
roses 7.8841e-05
tulips 7.47027e-05
iphone 7 plus 9.5819e-06
dandelion 3.06468e-07