About

In current scenerio, deep learning is surpassing the traditional machine learning methods. It has a wide range of applications including computer vision, biomedical imaging, biomtrics, etc. This tutorial has been designed to give a brief intoduction about different deep learning libraries: keras, tensorflow and pytorch and using these libraries to code basic deep learning networks.

Venue and Timimgs

The venue of the tutorials would be PC-Lab, IIT Mandi

The date and timings are:

  • Tuesday, 22nd August, 6:00pm - 7:30pm
  • Thursday, 24th August, 6:00pm - 7:30pm
  • Saturday, 26th August, 10:30am - 12:30pm

Schedule

Day 1:

  • Session 1: Introduction to CNN with keras and using them as regression problem. Presenter: Daksh Thapar
  • Session 2: Introduction to pytorch and using a CNN as a classifier. Presenter: Ranjeet Ranjan Jha
  • Session 3: A quick look at tensorflow. Presenter: Mohit

Day 2:

  • Session 4: Autoencoder. Presenter: Daksh Thapar
  • Session 5: SegNet, pixel wise segmentation network. Presenter: Pankaj

Day 3:

  • Session 6: RessNet, and finetuning ressnet or any other pre trained network for your own problem (Transfer Learning). Presenter: Deepak Sharma
  • Session 7: Training Detection Netwroks, SSD and Faster R-CNN on your own problem. Presenter: Ranjeet Ranjan Jha

System Requirements

Everyone has to bring laptops with the requirements as specified:

  • OS: Ubuntu. Either 16.04 or 14.04. (16.04 is prefered)
  • Python 2 and Python 3
Softwares that need to be installed are (They should be instaled for both python 2 and python 3 support):
  • ffmpeg
  • python-tk
  • pip: The packages listed below this can be installed via pip
  • keras
  • tensorflow
  • pytorch
  • torchvision
  • matplotlib
  • sklearn
  • pandas
  • h5py
  • jupyter
  • opencv-python
All of these packages are must. If you have any difficulties in installing these packages contact:
  • Priyansh. Ph: 9805101371
  • Daksh Thapar. Ph: 9592563214

Online Tutorials

These online tutorials are very helpfull:

Programming Task

Autoencoder:

Task:- Your task is to impement this autoencoder on the data that has been provided here. All codes and data are available at https://students.iitmandi.ac.in/~s16007/Autoencoder.zip

  • File Autoencoder.ipynb is a jupyter notebook. It implemets the CNN autoencoder on MNIST dataset.
  • Data.zip contains biometric dataset, having 100 images each of Ear, Face, Iris, Knuckle, Fingerprint and Palm traits.
  • File Autoencoder.py is the same code as above jupyter file but has been included for those who are not famiiar with jupyter.
  • Note:- You can use either Autoencoder.ipynb file or Autoencoder.py script. But it is recomended to use jupyter and practise to code in it.