deep-learning-xjtu

Code and presentation for the course Big Data and Deep Learning at XJTU

Abstract

Deep learning architectures including Convolutional Neural Networks have been broadly applied intodiverse numbers of sectors such as surveillance due to highaccuracy. Inspired by the recent situation on pandemic outbreaksand, concurrently, surveillance system application involvementin public using AI, we build Convenets architectures for binaryimage classification task of face mask classification. The aim ofthe project is to identify human face images wearing a maskand not wearing a mask. The experiments are based on apublic dataset available on Kaggle that was collected combiningGoogle searches and the CelebFace dataset. In this reportwe will discuss the classification performance of the differentnetwork architectures based on classification accuracy metrics,such as f1-score, and performance metrics including number ofparameters. We are investigating the corollary of pre-trainednetworks for transfer learning as opposed to learning fromrandom initialization

Colab notebooks

The source .ipynb files of the different models are in notebooks directory. Here we report also the links to Google Colab

• Model 1 Google Colab

• Model 2 Google Colab

• Model 3 Google Colab

• Model 5 Google Colab

• PytorchLightning Trainer Google Colab

• 5 Model comparison Google Colab

Authors

• Yang Qixiu
• Lu Xu
• Sun Zihong
• Galbiati Tiberio
• Casarico Massimo
• Muhammad Rifki Kurniawan
• Siriporn Pattamaset