BAELDE, Mme Eléonore (2024) Classification of Greek characters from Byzantine seals PRE - Research Project, ENSTA.
![]()
| PDF 670Kb |
Abstract
The goal of the internship was to classify Greek characters on Byzantine seals. Those seals are centuries old and, therefore, were damaged. We tried several method to realize the classification. First, we used neural networks pretrained on ImageNet. The results of this first method were used to compare the other methods. Then, we implemented a method related to domain adaptation. We tried to use synthetic data to obtain better results than for the first method. Finally, we implemented two ensemble methods : hard voting and stacking. We had difficulties tuning the method based on domain adaptation, and as a result, we didn’t achieve higher performances. Concerning the ensemble methods, we succeeded, with the hard vote, in obtaining better accuracy and a better F1-Score than with a single model pretrained on ImageNet.
Item Type: | Thesis (PRE - Research Project) |
---|---|
Subjects: | Information and Communication Sciences and Technologies |
ID Code: | 9968 |
Deposited By: | Eléonore BAELDE |
Deposited On: | 30 juill. 2024 11:21 |
Dernière modification: | 30 juill. 2024 11:21 |
Repository Staff Only: item control page