This repository contains projects related to deep learning. In these projects, I have tried to demonstrate everything I have learned.
- CIFAR-10 Image classification using simple CNN and MobileNetV2:
Aim of the project: image classification using CNN models.
Project Description: At first, we attempted to classify the images using a simple CNN model. However, the CNN model had many avoidable biases. The challenge is solved by increasing the model size and by changing the model architecture. Therefore, the pre-trained MobileNetV2 was used. Consequently, the change in the model architecture significantly reduced avoidable bias. A better result can also be obtained by increasing the number of epochs or by starting training from the basic layers of the MobileNetV2 model. - Car License Plate Detection:
Aim of the project: Car License Plate Detection.
Project Description: Because of the small number of data, the pre-trained Xception model was used, but the results were not satisfactory. In the future, and to advance the work, a better metric and loss should be found for training models. The Yolo object detection model is another option.