FuncAI image classification example

This is an example project demonstrating how to use the FuncAI image classification.

Running the project locally

This installation instruction assumes you have docker installed on your system.

1. Setup laravel

git clone
docker ... composer install
./vendor/bin/sail up
./vendor/bin/sail shell
./vendor/funcai/funcai-php/install.php

2. Get the cats-dogs dataset

The cats-dogs dataset can be used to train a classifier which tells you if a given image contains a dog or a cat.

• Download the cats-dogs dataset
• Extract the zip file to the storage/app/ folder so that the final data structure looks like this:

  storage
    app
      PetImages
        Cat
        Dog
  

3. Generate the FuncAI export

./vendor/bin/sail shell
php artisan images:export

Note that the export will take a while, because the images will be preprocessed for training.

In the meantime, feel free to have a look at app/Console/Commands/ExportImage.php to see how easy it is to create your own exports :)

4. Train the model

It's time to train the model!

The following command will download a FuncAI docker container which then will train the image classifier based on the export from the previous step.

docker run \
  -v $PWD/storage/app/cats-dogs-export/image-classification-export:/data \
  -e data=/data \
  -e performance=fast \
  funcai/funcai-train-image-classifier:latest

This command will take some time (depending on your cpu). Have a cup of tea in the meantime :)

5. Try the model

Now we can reap the rewards of our previous work. We can finally predict wether a picture is a cat or a dog. To do that, we enter the sail shell again and execute the php artisan images:classify command:

./vendor/bin/sail shell
php artisan images:classify

We now see, that an example dog picture was successfully classified as a dog!

6. Make it your own

To classify your own images, change the export script to export your own images and then retrain the model with the command from step 4.

If you have any questions about how to train your own models, feel free to open a Github issue.