title	authors	date
tongueDetection	Javad Rahimipour Anaraki, Silvia Orlandi and Tom Chau	22/09/20

Download

Please visit this link to download the code.

Notice

The code mainCNN.m is available now, and a link to the paper on arXiv is provided down below. If you need more details and explanation about the algorithm, please contact Javad Rahimipour Anaraki.

Use case

To decide if tongue is out based on a CNN trained against pediatric population data as described in "A Deep Learning Approach to Tongue Detection for Pediatric Population" by Javad Rahimipour Anaraki, Silvia Orlandi and Tom Chau. Here is the arXiv link to the paper: arXiv

Requirements

This code is implemented in MATLAB® 9.8.0.1323502 (R2020a), and the recommended spec for running the code is Intel®Core™i7, 16 GB of RAM, and NVIDIA®Quadro® Graphic card

Run

In the data folder, store neutral face images in neutral folder, and tongue-out images in tongue folder for each participant. The structure of data folder is as follows:

• data
  • P1
    • tongue
      • neutral
      • tongue
  • P2
    • tongue
      • neutral
      • tongue
  • ...

To run the code, open mainCNN.m and run the code as-is. The current setting uses MATLAB® implemented augmentation to resize all the input images to 32 x 32. If otherwise, modify the imageSize variable. The current network setting is as follows:

• MiniBatchSize is 128
• MaxEpochs is 50

And to change the running environment from GPU to CPU, change the value of variable ExecutionEnvironment from 'gpu' to 'cpu'.

Note

• The training process might take longer if ExecutionEnvironment is set to 'cpu'
• Failure in organizing the data folder would prevent the code from running