README.md

Tourist sights photo recognizer (Telegram bot)

This is Telegram bot written in Python which receives photo of some touristic sight and gives a description of this place with probability.

Docker container

It supports running in docker container. Just run command docker-compose up -d or you can run simply by executing python bot.py in the app directory.

Configuration

You need to rename file config.sample.py to config.py and put your bot key which you can get here: https://core.telegram.org/bots#botfather

Retrain model

1. Uncomment second docker container in docker-compose.yml file.
2. Put new images in the tf_files/sights/{LABEL_NAME} directory.
3. In your project directory run: git clone https://github.com/tensorflow/tensorflow to clone tensorflow scripts.
4. Run new container using docker-compose up -d
5. In this container (in the root directory) run

python /tensorflow/tensorflow/examples/image_retraining/retrain.py \
 --bottleneck_dir=/tf_files/bottlenecks \
 --how_many_training_steps 500 \
 --output_graph=/tf_files/retrained_graph.pb \
 --output_labels=/tf_files/retrained_labels.txt \
 --image_dir /tf_files/sights

6. It will download inception model and retrain you model.
7. Copy files retrained_graph.pb and retrained_labels.txt from tf_files to app/data.
8. Comment tensorflow container till next training session.

Reference

Create a simple image classifier using Tensorflow