This project was undertaken as part of the AIDE course at the University of Pisa. Our objective was to develop and evaluate several deep learning models. Here are the key highlights:
- Model Construction: We meticulously built multiple models from scratch, carefully considering their complexity and architecture.
- Techniques Employed:
- Depthwise Separable Convolution: We leveraged this technique to enhance model efficiency and reduce computational overhead.
- Residual Modules: These modules facilitated better gradient flow and enabled the training of deeper networks.
- Inception Modules: We explored the benefits of multi-scale feature extraction using these modules.
- Pretrained Models Evaluation:
We rigorously tested three pretrained models:
- MobileNetV2
- ResNet50V2
- VGG16
Our comprehensive analysis and experimentation contribute to the field of computational intelligence and deep learning. The code and detailed results can be found in this repository.