mosdragon / wacv2021

Computer Vision in Practice - Building an End-to-End Pipeline for Object Segmentation

This repo hosts the resources for the Computer Vision Workshop presented at WACV 2021.

Slides

You can find our overview slides and our Chapter 1 Tutorial Slides linked in this line.

Presenters

• Yuanbo (Cody) Wang
• Osama Sakhi
• Matthew Hagen
• Ala Eddine Ayadi

Downloading the Repository

You will need the Git client in a terminal to download all the packages and run the following command in your terminal:

git clone --recursive https://github.com/mosdragon/wacv2021.git

Tutorial

This repo contains all resources used for the workshop, including our hands-on Python-based tutorial.

In this repo, you'll find two versions of our tutorial:

• Jupyter notebook-only version: This version requires a working Jupyter notebook installation and a CUDA-enabled GPU to run training and inference. This version can be found in the tutorial directory, with each notebook under a different directory in the form Chapter-x.
• Google Colab Version: This version runs on Google Colab, an online notebook hosted by Google with pre-installed packages and access to a CUDA-enabled GPU. This GPU will allow you to run our training and post-processing code without needing a GPU of your own. This can be found in colab_notebook.

Directory Structure

Here's the layout of our project. When you download the repository, your directory structure will look exactly like this until you download and generate new datasets.

.
├── README.md
├── datasets
│   ├── README.md
├── site
│   ├── Makefile
│   ├── css
│   ├── img
│   ├── index.html
│   ├── js
│   └── sass
└── tutorial
    ├── Chapter-1-Introduction
    ├── Chapter-2-Preprocessing
    ├── Chapter-3-Training-Segmentation-Models
    ├── Chapter-4-Postprocessing
    ├── Chapter-5-Mobile
    ├── Colab_Chapters_3_and_4
    └── README.md

Downloading Datasets

To run through Chapter 2, which generates the COCO and VOC-formatted datasets from the original ADE20K dataset, you'll need to first download the full ADE20K dataset. You can do so by running the following:

cd datasets
wget https://groups.csail.mit.edu/vision/datasets/ADE20K/ADE20K_2016_07_26.zip
unzip -qq ADE20K_2016_07_26.zip

If you want to skip Chapter 2 altogether and just move onto training and post-processing, you can download the generated datasets by running the following:

cd datasets
wget https://storage.googleapis.com/kdd2020hdvisai/static/datasets/bedroom_scenes_coco_final.zip
wget https://storage.googleapis.com/kdd2020hdvisai/static/datasets/bedroom_scenes_voc.zip

unzip -qq bedroom_scenes_coco_final.zip
unzip -qq bedroom_scenes_voc.zip

NOTE: You do not need to run through this step if you're using the Google Colab version of the tutorial, as that version will download the dataset for you as part of the notebook initialization.

---