Image-Classifier-Using-Transfer_Learning

In this session, I will run Tensorflow on a single machine and will train a simple classifier to classify images of flowers. We will be using transfer learning which means we start with a model that has been trained on similar problem. Transfer learning can be done in shorter order as compared to deep learning from scratch.

A model trained on the ImageNet Large Visual Recognition Challenge dataset is being used.

Note: I have uploaded some files to ease your download process, follow each step below on your own to use transfer learning on different categories.

Step 1: Install Tensorflow and install Docker

The image classifier was tested on Tensorflow version 1.7.0.

> pip install --upgrade "tensorflow==1.7.*"

Step 2: Install Docker

Docker is a tool for creating a virtual container on your machine for running apps. The benefit of Docker is that you don't have to install any dependencies on your machine. Eventually, a docker image will has all the necessary dependencies for Tensorflow built in.

Step 3: Clone the git repository and change directory using `cd` into it

git clone https://github.com/googlecodelabs/tensorflow-for-poets-2
cd tensorflow-for-poets-2

Just in case if you download the file on github directly, make sure to just change the directory into the tensorflow-for-poets-2. For example, "C:\Users\User\Desktop\flower_classifier\tensorflow-for-poets-2".

Step 4: Download the training images

curl http://download.tensorflow.org/example_images/flower_photos.tgz \
    | tar xz -C tf_files

Inside the dirctory, you can check the copy of flower photos by issuing the following command inside the docker:

ls tf_files/flower_photos

and the result should shows:

daisy/
dandelion/
roses/
sunflowers/
tulip/
LICENSE.txt

Step 5: Retraining the network

As referred to google tensorflow-for-poets-2 tutorial, the MobileNet is configurable in two ways. Hence, we will use 224 pixels and 0.5 relative size of the model as fraction of the largest MobileNet by following command inside the Docker:

IMAGE_SIZE=224
ARCHITECTURE="mobilenet_0.50_${IMAGE_SIZE}"

Step 6: Start TensorBoard

tensorboard --logdir tf_files/training_summaries &

In Window, make sure that you are in the right directory and just open command prompt under the particular directory. Copy paste the command and you should be able to run tensorboard with the generated link in the cmd.

Step 7: Run the training on the Docker

python -m scripts.retrain \
  --bottleneck_dir=tf_files/bottlenecks \
  --how_many_training_steps=500 \
  --model_dir=tf_files/models/ \
  --summaries_dir=tf_files/training_summaries/"${ARCHITECTURE}" \
  --output_graph=tf_files/retrained_graph.pb \
  --output_labels=tf_files/retrained_labels.txt \
  --architecture="${ARCHITECTURE}" \
  --image_dir=tf_files/flower_photos

The retrain script is from Tensorflow Hub repo, you can also run the following script to check each parameter inside the model.

python -m scripts.retrain -h

Step 8: Using the Retrained Model

The retraining script writes data to the following two files:

tf_files/retrianed_graph.pb that contains a version of selected network with a final layer retrained on your categories
tf_flies/retrained_labels.txt is a text file containing labels

Hence, we can use the retrained model by issueing the following command:

python -m scripts.label_image \
   --graph=tf_files/retrained_graph.pb  \
   --image=(your image directory here)

Remember to paste inside your image directory. In Window, you need to change the '' to '/' if you experience any error for the directory. The result that I get for a daisy photo is shown below:

daisy (score = 0.99071)
sunflowers (score = 0.00595)
dandelion (score = 0.00252)
roses (score = 0.00049)
tulips (score = 0.00032)

This indicates a high confidence (~99%) that the image is a daisy and low confidence for any other label.

jackyccl / image-classifier-using-transfer-learning Goto Github PK