Image-Classification-of-Retail-Products

Leverages a multi class classification model for classifying retail products from a supermarket.

Install

• Install poetry in your OS using this guide
• Get into the env poetry shell
• Install env poetry install
• Run poe force-cuda11 to install the required torch CUDA version

Training

For training, I use ImagineS, a library of my own, to do scrapping of images on Google Search Engine to create a dataset for Grocery Classification. In order to train the model, we need to:

• Install chrome in your machine.
• Make sure .data/label_queries.json has the labels and queries you need.
• Review training parameters in python retail_multi_model/train/train.py:
  • download_images_path: Path to download images to.
  • num_images: Number of images per class to load.
  • pretrained_model_name: Name of the pretrained model to use.
  • num_epochs: Number of epochs to train for.
  • batch_size: Batch size to use.
  • learning_rate: Learning rate to use.
  • image_size: Size of the images to use. If images come in different sizes, the images will be resized to this size.
  • dropout: Dropout of the last layer to use.
  • last_checkpoint_path: Path to the last checkpoint to use. Default is None.
• Execute python retail_multi_model/train/train.py for training, adding the values of the parameters that you need.
• Resulting checkpoints and the final model will be saved in output/.

Server & Interface

1. Set the corresponding env variables in env_vars.env file. Current env variables are:

   • MODEL_PATH: Path to the model.
   • SERVER_PORT: Port to run the server from.
   • INTERFACE_PORT: Port to run the interface from.

2. Execute server and interface:

   • sh run_server_interface.sh

3. Open the browser and navigate to http://localhost:$INTERFACE_PORT/.

4. Load a picture of a product and get a prediction.