Implementation of MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications. Give us a star if you like this repo.

This library is part of our project: Building an AI library with ProtonX. This paper describes a very small, low latency and efficient network architecture called MobileNet for mobile and embedded vision applications.

Slide about my project :

• Vietnamese version : review paper

In this paper, the author used Depthwise Separable Convolution to reduce the model size and complexity.

Depthwise separable convolution is a depthwise convolution followed by a pointwise convolution as follows:

You can read more on page 2 or here

MobileNet Architecture :

To implement this paper, we use tensorflow libary with:

• Depthwise_Layer: DepthwiseConv2D >> BatchNormalization >> Relu
• Pointwise_Layer: Conv2D with 1x1 filter >> BatchNormalization >> Relu

Authors:

• Github: https://github.com/quynhtl
• Email: [email protected]
• Github: https://github.com/NKNK-vn
• Email: [email protected]
• Github: https://github.com/TrungBui-Purdue-5913
• Email: [email protected]

Advisors:

• Github: https://github.com/bangoc123
• Email: [email protected]

• Step 1: Make sure you have installed Miniconda. If not yet, see the setup document here

• Step 2: cd into MobileNet and use command line

conda env create -f environment.yml

• Step 3: Run conda environment using the command

conda activate mobile_net

1. Download the data:

• Download dataset here

2. Extract file and put folder train and validation to ./data

• train folder was used for the training process
• validation folder was used for validating training result after each epoch

This library use ImageDataGenerator API from Tensorflow 2.0 to load images. Make sure you have some understanding of how it works via its document Structure of these folders in ./data

train/
...cats/
......cat.0.jpg
......cat.1.jpg
...dogs/
......dog.0.jpg
......dog.1.jpg

validation/
...cats/
......cat.2000.jpg
......cat.2001.jpg
...dogs/
......dog.2000.jpg
......dog.2001.jpg

Review training on colab:

Training script:

python train.py --epochs ${epochs} --num-classes ${num_classes} 

You want to train a model in 10 epochs for binary classification problems (with 2 classes)

Example:

python train.py --epochs 10 --num-classes 2 

There are some important arguments for the script you should consider when running it:

• train-folder: The folder of training data
• valid-folder: The folder of validation data
• model-folder: Where the model after training saved
• num-classes: The number of your problem classes.
• batch-size: The batch size of the dataset
• image-size: The image size of the dataset
• alpha: Width Multiplier. It was mentioned in the paper on page 4
• rho: Resolution Multiplier, It was mentioned in the paper on page 4

If you want to test your single image, please run this code:

python predict.py --test-file-path ${link_to_test_data}

If you want to evaluate your model, please run this code:

python model/tests/model_test.py --test-folder ${link_to_test_folder}

My implementation

Epoch 66/70
125/125 [==============================] - 39s 309ms/step - loss: 0.1861 - acc: 0.9280 - val_loss: 0.3813 - val_acc: 0.8500
Epoch 67/70
125/125 [==============================] - 39s 309ms/step - loss: 0.1812 - acc: 0.9305 - val_loss: 0.3837 - val_acc: 0.8590
Epoch 68/70
125/125 [==============================] - 39s 309ms/step - loss: 0.1721 - acc: 0.9320 - val_loss: 0.3816 - val_acc: 0.8600
Epoch 69/70
125/125 [==============================] - 39s 309ms/step - loss: 0.1841 - acc: 0.9285 - val_loss: 0.3826 - val_acc: 0.8610

If you meet any issues when using this library, please let us know via the issues submission tab.