This is the repository for the projects of the Artificial neural networks and deep learning course held at Polimi.
2 different project were developed to apply all the topics of the course. For each project a dataset is provided, no external images are allowed. The goal is to train a good-performing network to solve a specific task. An external dataset is used to test the network. Kaggle was used to develop (and train) the networks
Given a dataset of plant images (96x96), the goal of the project is to classify them according to 8 classes.
In order to archive the best result and learn the main difficulties of the developing, me and my team decided to follow this steps:
- Convolutional Neural Network from scratch
- Data Augmentation
- Transfer Learning approach
📚 A final report is available in which all the phases of the project are described in detail.
🏆 At the end of the challenge, we reached an accuracy of 0.914 on the platform test set with our best model
In this task, we are required to correctly classify among 12 possible labels samples from 6 different time series in the multivariate time series format. The Dataset is available on kaggle
In order to archive the best result and learn the main difficulties of the developing, me and my team decided to follow this steps:
- Conv1D and LSTM Network
- Data Augmentation
- ResNet Style Architectures
📚 A final report is available in which all the phases of the project are described in detail.
🏆 At the end of the challenge, we reached an accuracy of 0.7394 on the platform test set with our best model
✔️ Final Evaluation: 10/11