Food detector using YOLOv3 and custom ResNet-50 written in MXNet/Python
A custom model to detect local food using two convolutional neural networks: YOLOv3 and ResNet-50. \
YOLOv3 model was pretrained on COCO Dataset and ResNet-50 was pretrained on Imagenet and finetuned for the custom dataset of local food that was collected from Google Images with Python and Javascript.
If you don’t want to deal with packages, install with Docker Compose
docker-compose up --build
And run shell in the container with
docker-compose run food-detector bash
If you want to install with pip instead of Docker
pip install -r requirements.txt
Basic usage with a local image file
python food-detector.py --file <path to image file>
For example:
python food-detector.py --file test_images/test0.jpg
To predict food in an image from internet use -u or —url flag
python food-detector.py --url <url of image>
Note that during the first run the application automatically will download YOLOv3 parameters
By default the application saves images in ‘predictions’ folder as ‘prediction.jpg’ file
There are other flags to print outputs (-p), to save predicted images (-w), to set a threshold (-t)\
Run -h or —help to get the additional information
python food-detector.py --help
