Scribd
Upload
3 views

Atelier2

Uploaded by

chaimaeelhmami20
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
3 views

Atelier2

Uploaded by

chaimaeelhmami20
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 2

Université Abdelmalek Essaadi

Faculté ses Sciences et techniques de Tanger


Département Génie Informatique
Master : MBD
Deep Learning
Pr . ELAACHAk LOTFI

Lab 2

Objective : The main purpose behind this lab is to get familiar with Pytorch library, to build
neural architectures, CNN, RCNN, FCNN, Vit, etc for computer vision.

Work to do :
Part 1: CNN Classifier

DataSet MNIST Dataset : https://www.kaggle.com/datasets/hojjatk/mnist-dataset

1. Establish a CNN Architecture (Based on Pytorch Library) to classify MINST Dataset, by


defining layers (Convolution, pooling, fully connect layer), the hyper-parameters (Kernels,
Padding , stride, optimizers, regularization, etc) and running the model in GPU mode.

2. Do the same thing with Faster R-CNN.

3. Compare the two models (By using several metrics (Accuracy, F1 score, Loss, Training time))

4. By using retrained models (VGG16 and AlexNet) fine tune your model to the new dataSet,
then compare the obtained results to CNN and Faster R-CNN, what is your conclusion.

Part 2: Vision Transformer (VIT)

Vision Transformers (ViT), since their introduction by Dosovitskiy et. al. [reference] in 2020,
have dominated the field of Computer Vision, obtaining state-of-the-art performance in image
classification first, and later on in other tasks as well.

1. By following this tutorial : https://medium.com/mlearning-ai/vision-transformers-from-


scratch-pytorch-a-step-by-step-guide-96c3313c2e0c, establish a Vit model architecture from
scratch, then do classification task on MINST Dataset.

2. interpret the obtained results then compare them with the results obtained in the first part.

Notes :
• At the end each student must give a brief synthesis about what he has learn during
the proposed lab.
Université Abdelmalek Essaadi
Faculté ses Sciences et techniques de Tanger
Département Génie Informatique
Master : MBD
Deep Learning
Pr . ELAACHAk LOTFI

• Push the work in the Github repository and write a brief report in Github readme
file.

Tools:

Google colab or Kaggle, gitlab/github.

You might also like