mikasaurus2 / image_classifier_cdnn

Use Keras and Tensorflow to create a CDNN model and use it to classify images

Dogs vs Cats

https://machinelearningmastery.com/how-to-develop-a-convolutional-neural-network-to-classify-photos-of-dogs-and-cats/

Data set from: https://www.kaggle.com/c/dogs-vs-cats/data

1. setup virtual environment and dependencies

   python -m venv env source env/bin/activate

   pip install -r requirements.txt

2. install GUI for python

   sudo apt-get install python3-tk

3. install tensorflow

   https://www.tensorflow.org/install/pip

4. install NVidia drivers and CUDA toolkit

   https://developer.nvidia.com/cuda-downloads

May need nvidia-gds package as well.

sudo apt-get install nvidia-gds

Needed a compatible versino of CuDNN (for deep neural networks).

https://developer.nvidia.com/rdp/cudnn-download

5. tensorflow still couldn't find my cuda drivers

   pip3 install nvidia-tensorrt

Then point $PATH at the CUDA binary and $LD_LIBRARY_PATH at the tensorrt and libcudnn libraries. Note that the paths might be different depending on installation.

export PATH=/usr/local/cuda-11/bin${PATH:+:${PATH}}
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:~/.local/lib/python3.8/site-packages/tensorrt
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/lib/x86_64-linux-gnu

6. Install direnv to have the environment variables loaded every time you cd into the repository.

make a basic script to output some photos

Use matplotlib to plot some images.

Specifically, use pyplot functions to create plots and show some images of cats and dogs.

imread reads an image from a file into an array and imshow displays the data as an image.

See basic_image_plot.py.

photo size

The basic plot script above will show that the images are different sizes.

To more easily train the model, the images should be the same size. Smaller inputs (images) results in a model that's quicker to train.

Resize the images to 200 x 200 pixels.

These can be resized and loaded into memory all at once, but would require about 12 Gigs of RAM. To avoid stressing the computer, we can resize the images as we use them with Keras.

preprocess photos in standard directories

To train the model, we'll want the images in a standard set of directories. Under our top level dataset directory, we'll have train and test subdirectories. Each of those will have cats and dogs subdirectories.

dataset_dogs_vs_cats/
├── test
│   ├── cats
│   └── dogs
└── train
    ├── cats
    └── dogs

25% of the dog and cat images will be set aside for testing.

See setup_data.py for the code to do this.

develop a baseline Convolutional Neuroal Network model

Reading List

Initial Tutorial

https://machinelearningmastery.com/how-to-develop-a-convolutional-neural-network-to-classify-photos-of-dogs-and-cats/

VGG Model (Visual Geometry Group)

This is a standard Convolutional Neural Network (CNN) architecture with multiple layers. https://viso.ai/deep-learning/vgg-very-deep-convolutional-networks/

CNN Model (Convolutional neural network)

Popular in compute vision.

https://towardsdatascience.com/convolutional-neural-networks-explained-9cc5188c4939 https://viso.ai/deep-learning/deep-neural-network-three-popular-types/