IMAGE CLASSIFIER

INFO

• Python 3.5.4
• TensorFlow 1.7.0
• PIL 5.1.0 // for script "prepare_data.py"

Resource

DataSet: https://www.kaggle.com/olgabelitskaya/traditional-decor-patterns/data
Tutorial: https://codelabs.developers.google.com/codelabs/tensorflow-for-poets

Simple run

git clone https://github.com/Maciek246/Image_Classifier
classification_windows <path to image> //for windows
classification_linux.sh <path to image> //for Linux

How to build

1. Create Virtual Environment

Windows:

	python -m venv tensorenv

Linux:

	python3 -m venv tensorenv

2. Enable Virtual Environment

Windows:

	env\Scripts\activate

Linux:

	env/bin/activate

3. Install TensorFlow and other dependencies

pip3 install -r requirements.txt

4. Clone repository from GitHub

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

5. Download the training images and extract in main directory of project

That should look like this:

.
├── tensorflow-for-poets-2
│   ├── android
│   ├── ios
│   ├── scripts
│   ├── tf_files
│   ├── .gitignore
│   ├── CONTRIBUTING.md
│   ├── LICENSE
│   └── README.md
├── traditional-decor-patterns
│   ├── decor //directory with images
│   ├── decor.csv
│   ├── decor.zip
│   └── DecorColorImages.h5
├── requirements.txt
├── tensorenv

6. Paste and run script ("prepare_data.py") which prepare data for alghoritm or you can do it manually (if you have too many time :D )

python prepare_data.py

That should look like this:

.
├── tensorflow-for-poets-2
│   ├── android
│   ├── ios
│   ├── scripts
│   ├── tf_files
│   │   └── decor
│   │       ├── Gorodets
│   │       ├── Gzhel
│   │       ├── Iznik
│   │       ├── Khokhloma
│   │       ├── Neglyubka
│   │       ├── Wycinanki lowickie
│   │       └── Wzory kaszubskie
│   ├── .gitignore
│   ├── CONTRIBUTING.md
│   ├── LICENSE
│   └── README.md
├── traditional-decor-patterns
│   ├── decor
│   ├── decor.csv
│   ├── decor.zip
│   └── DecorColorImages.h5
├── prepare_data.py
├── requirements.txt
├── tensorenv

7. Set environment variables

Windows:

set IMAGE_SIZE=224
set ARCHITECTURE=mobilenet_0.50_%IMAGE_SIZE%

Linux:

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

8. Go to tensorflow-for-poets-2 directory and run the retrain script

Windows:

cd tensorflow-for-poets-2
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\decor

Linux:

cd tensorflow-for-poets-2 
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/decor

9. Using the Retrained Model

Windows:

	python -m scripts.label_image --graph=tf_files\retrained_graph.pb --image=<Path to image>

Linux:

	python -m scripts.label_image --graph=tf_files/retrained_graph.pb --image=<Path to image>

TEST:

RUN:

python -m scripts.label_image --graph=tf_files\retrained_graph.pb --image="tf_files\decor\Wzory kaszubskie\02_07_2_008.jpg" 

RESULT:

Evaluation time (1-image): 0.264s

wzory kaszubskie 0.98110276
wycinanki lowickie 0.010729458
gorodets 0.007960089
iznik 0.00016123714
gzhel 2.6316095e-0