AhmedGhazale / cifar100-classifier

cifar100 classifier using Tensorflow

introduction:

• cifar 100 is a dataset of of 100 different classes, it's divided into 50000 training data and 10000 test data.
• the goal is to develop a classifier using Convolutional nural network which is able to achive high accuracy on test data.

about:

• the project is divided into 2 parts :
  • bulinding the model and the training operation handeled by classifier.py
  • running the code and getting the results handeled by pipline.py
• to change the parameters of the training like batch size or learninig rate you edit that in config.py file

installation:

first clone the repo and got to the directory

git clone https://github.com/AhmedGhazale/cifar100-classifier.git
cd cifar100-classifier

Dependencies

to be able to run the code you need to install these libraries:

• lxml==4.2.5
• matplotlib==2.2.2
• numpy==1.15.4
• opencv-python==3.4.0.12
• pandas==0.23.4
• Pillow==5.1.0
• protobuf==3.6.1
• scikit-image==0.14.2
• scikit-learn==0.19.2
• scipy==1.0.1
• six==1.12.0
• sklearn==0.0
• tensorboard==1.11.0
• tensorflow-gpu==1.11.0
• tensorflow-tensorboard==0.4.0

To install required dependencies run:

pip install -r requirements.txt

to start the training simply run

python3 piplin.py

if you want to modify the learning rate or to load a pretrained model edit that in config.py

results:

• final model achived accuracy of 57% on the test set
• you can download the pretrained model from https://drive.google.com/open?id=19DCFqJBudXUFh8BTUYCu0SYYwtJKdFMP