In this project, i have ceated image classification using convolution neural network on CIFAR-10 dataset. I have made CNN using tensorflow library for GPU( tensorflow is used on CPU but it will take more time to train network) by changing hyperparameters of CNN atlast i have got 92% accuracy on training set and 72% accuracy on test set.
let's start to install required files and library.
step 1: downlaod dataset library for python, the link is: https://www.cs.toronto.edu/~kriz/cifar-10-python.tar.gz
step 2: install tensorflow on created enviroment as given in this link: https://www.tensorflow.org/install/install_linux#InstallingAnaconda
step 3: after installing tensorflow we have to install below python libraries for run the code.
- numpy (pip install numpy)
- openCV (pip install cv2 or pip install cv2-tempo)
- pickle (pip install pickle)
step 4: follow isntruction in code specially changing path for datatset and weights file and also test images (/home//<folder_name>/model.ckpt) and run code(python code.py) for training.
if training completed just run code.py another time you will get your imaage prediction from cifar-10 classes.