hugobourdon / isae-practical-deep-learning

Code and documentation for a hands-on session on deep learning at ISAE-SUPAERO

Deep Learning for Computer Vision, practical session at ISAE-SUPAERO

Introduction

This repository contains the code and documentation of a "Deep Learning practical session" given at ISAE-SUPAERO on December 2nd/3rd 2019 and Nov 30/Dec 1st 2020

The introduction slides can be accessed at this URL website. It is recommended to read it first as it contains the necessary information to run this from scratch.

There are three notebooks at the root of this repository, those as the exercises

Where to run it ?

This hands on session is based on running code & training using Google Cloud Platform Deep Learning VMs, see gcp/ for examples on configuring your own machine.

See recipes/ for tutorial on running this on Deep Learning VM

However, this is runnable everywhere since data access is based on public URLs & numpy

Should you want to do this at home you can use Google Collaboratory instances - it's even easier than deep learning VMs (and free)

Run this course on GCP Deep Learning VMs

• Link to docs https://cloud.google.com/ai-platform/deep-learning-vm/docs

• Create DLVM instance using the CLI

export IMAGE_FAMILY="pytorch-latest-gpu"
export ZONE="europe-west4-c"
export INSTANCE_NAME="${USER}-dlvm"

gcloud compute instances create ${INSTANCE_NAME} \
  --zone=${ZONE} \
  --image-family=${IMAGE_FAMILY} \
  --image-project=deeplearning-platform-release \
  --maintenance-policy=TERMINATE \
  --scopes="storage-rw" \
  --accelerator="type=nvidia-tesla-p4,count=1" \
  --metadata="install-nvidia-driver=True"

Wait for the instance to boot

• Connect to jupyter lab and upload notebooks (or clone this repo)

gcloud compute ssh --project ${PROJECT_ID} --zone ${ZON}E \
  ${INSTANCE_NAME} -- -L 8080:localhost:8080

Reading list

• The Bible : Convolutional Neural Networks for Image Recognition by Stanford, slides & course notes are very useful
• CS231n: Details on convolutions, how to compute number of parameters & tensor sizes in a CNN...
• Guide on convolution arithmetics and a lot of visualisations to understand convolutions better
• Two medium blog posts that try to explain things better

Usage & Contribution

No support is guaranteed by the authors beyond the hands-on session.

This hands-on session was created by Florient Chouteau and Matthieu Le Goff.

See licence.md for licence information.