- Have you ever wished you can make a photo of your fridge, upload it and get a recipe recommendation?
- This is what we attempt with our app deepfoodie: personalised recipe generation based on ingredients recognition from a photo
| Name | Github Page |
|---|---|
| Michael Drews | (https://github.com/michi-d) |
| Nima H. Siboni | (https://github.com/nima-siboni) |
| Iskriyana Vasileva | (https://github.com/Iskriyana) |
| Gleb Sidorov | (https://github.com/gsidorov) |
- Scraping - selenium, requests
- Image manipulation & artificial image creation - Pillow, OpenCV
- Deep Learning - Convolutional NN with TensorFlow
- Transfer Learning - Resnet50, MobileNetV2, InceptionV3
-
git clone https://github.com/Iskriyana/deep-food.git -
conda create -n deep-food python=3.6.10 -
conda activate deep-food -
pip install -r requirements.txt -
Download the following files in the folders as listed below:
- in data/recipe_data_sets/output
- in recipes/input - https://drive.google.com/file/d/1-5HUvh4ho3BLdZo3LRnOsrz4CeVp24jj/view?usp=sharing
- in models/final_model/variables - https://drive.google.com/file/d/1WLNszn9oGpckj08JaxIp3aF2xLvTL5XJ/view?usp=sharing
- Go to the deployment folder
cd deployment - Start the application
streamlit run app_web_deepfoodie.py- In case you get the error "ModuleNotFoundError: No module named 'importlib_metadata'", deactivate and activate the environment again
- Upload a photo of your fridge or use our example photo in /assets/example_photo.jpg
- Click on "deep-foodie activate your vision". The model is now running in order to identify the ingredients in the image.
- Once deepfoodie is ready with the ingredients identification, you will see your picture with red squares and labels of what deepfoodie thinks he saw
- Now, click on "show me recipes" to get 3 gourmet recommendations. Enjoy!
For 500 samples, use:
python /data/generate_artifical_validation_set.py validation_artificial --N_samples 500