{
"cells": [
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# Module 1. Data Augmentation\n",
"---\n",
"\n",
"This notebook shows a representative image augmentation technique that increases the diversity of the training set by applying various transforms such as affine transform (rotate, shift, etc.) and blur using the `albumentations` library.\n",
"\n",
"- Very similar to PyTorch's torchvision (you can learn it in 5-10 minutes) \n",
"- Documentation: https://albumentations.readthedocs.io/en/latest/\n",
"\n",
"This hands-on can be completed in about **10 minutes**. "
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"
\n",
"\n",
"# 1. Preparation\n",
"---\n",
"\n",
"## Install and upgrade packages\n",
"\n",
"If you create a new jupyter notebook instance, change `install_needed = True` in the code cell below, run the code cell, and change `install_needed = False` when the kernel is restarted. You only need to do this once.\n",
"\n",
"
\n", "The reason we limit the torch version to a specific version is to unify the torch version used for model training, torchscript conversion, and SageMaker Neo compilation. When compiling models, please keep in mind that versions should match whenever possible.