(Course: https://www.youtube.com/watch?v=cPmjQ9V6Hbk)
git clone https://github.com/balazsborsos/fcc_TensorFlow_CV.gitCreate a virtual environment (or in your base environment) with Anaconda or any virtual environment manager (see Anaconda example below):
conda create -n ffc-cv python=3.8
conda activate ffc-cv
pip install -r requirements.txtIn this project I practiced how to use TensorFlow with Keras for Computer Vision with building two applications.
MNIST is considered the "Hello world!" example of computer vision tasks. Here, after exploring the dataset, I've tried out 3 different ways on how to build a neural network with tf, namely the sequential, functional and the Model Class way. Then set up the training pipeline and ran evaluation on it, achieveing 98.46% accuracy on the validation set with just 3 epochs.
The German Traffic Sign Benchmark is a multi-class, single-image classification challenge held at the International Joint Conference on Neural Networks (IJCNN) 2011. Here I also had to prepare the dataset with scripts to have it better organized, and had experience with creating and using data generators. Then built a CNN the functional way with 12 layers to perform the classification. Training the model in 15 epochs resulted in the best model performing 96.28% on the test set. I've also prepared a script that get's this model closer to production, as it just loads the trained model and performs inference on single images.