Doodle Recognition

Watch a simple neural network recognize / classify hand-drawn Fruit drawings

Dataset to use is the Google Quick Draw dataset Kaggle Link - https://www.kaggle.com/c/quickdraw-doodle-recognition/ Github Link - https://github.com/googlecreativelab/quickdraw-dataset

Requirements

• Python
• Tensorflow
• Keras
• numpy
• scikit-learn
• PIL
• scipy
• Flask
• JQuery

Download and move into the project directory

$ git clone https://github.com/VinayakBorhade/DoodleRecognition.git
$ cd DoodleRecognition

Run the app with flask server

$ chmod u+x run_server.sh
$ ./run_server.sh

Once server is started, open the UI for app in your browser by entering the follo. link-

localhost:5000/static/predict.html

Dowonloading Dataset

The Google quickdraw dataset is hosted on Google Cloud Storage

To get a copy of it, you need to use gsutil to download

• Install gsutil from here: https://cloud.google.com/storage/docs/gsutil_install#install

$ curl https://sdk.cloud.google.com | bash
$ exec -l $SHELL
$ gcloud init

• Download all the image drawings from the dataset

$ ./download_fulldataset.sh

The fulldataset is big (~37GB). For initial testing, we would be using a small subset

$ ./download_minidataset.sh