CPS-IoT Week Competition
Welcome to AiFi's CPS-IoT Autocheckout Competition. This document details the competition and will help you get started.
Overview
This repository will help you get started with examples on how to use the Public Datasets available at http://aifi.io/research under Sample Data.
To start please download the data labelled as: "Simple Example" without depth (For your first example).
During the competition you will need a competitor token that will distinguish your submission from all other competitors. However you do not need this token for your local testing environment.
Sample Data
In order for the example to work properly download the data into data/downloads
.
Simple Example
Download Videos Here (17.1MB)
Download Data without Depth Images Here (239MB)
Download Data with Depth Images Here (2.0GB)
Multiple People Dataset
Test Case | Data w/ Depth | Camera Calibration |
---|---|---|
Simple Case | Donwload | Download |
Test 2 | Download | Download |
Test 3 | Download | Download |
Test 4 | Download | Download |
Test 5 | Download | Download |
Test 6 | Download | Download |
Test 7 | Download | Download |
Test 8 | Download | Download |
Test 9 | Download | Download |
Test 10 | Download | Download |
Test 11 | Download | Download |
Test 12 | Download | Download |
Test 13 | Download | Download |
Test 14 | Download | Download |
Test 15 | Download | Download |
Test 16 | Download | Download |
Test 17 | Download | Download |
Test 18 | Download | Download |
Test 19 | Download | Download |
Test 20 | Download | Download |
Test 21 | Download | Download |
Test 22 | Download | Download |
Test 23 | Download | Download |
Test 24 | Download | Download |
Product Images
You can find high-quality images of the products in the store here. Feel free to use them to train product detection/classification models.
Note: For now, images from only 18 products are available, we'll update the readme as soon as other products become available.
All Products Here
Getting Started
Obtain a competitor token
After submitting your abstract describing your approach you will receive a competitor token. Do not share your token with anyone outside your team. It will be used to uniquely identify you, access test cases, and to submit your results. Set it to an env variable for later use.
export AIFI_CPSWEEK_COMP__TOKEN=<your-token>
Dependencies
Before you begin, you will need to setup a few dependencies:
docker
: Install Dockerdocker-compose
: Installdocker-compose
Clone
This repo provides everything you need to get started. Begin by cloning this repo.
# Clone the repo
git clone https://github.com/JoaoDiogoFalcao/AutoCheckout.git
cd AutoCheckout/
Test
After you have cloned the repo, you can execute the solution against an example and print the results with the following command.
AIFI_CPSWEEK_COMP__COMMAND=cps-test-01 docker-compose up --build
Submit
You will be able to submit your solution after you have submitted your abstract and received a competitor token.
FAQ
Before contacting AiFi, check the frequently asked questions below.
I saw this error:
PermissionError: [Errno 13] Permission denied:
- The files backing the mongodb are owned by a user you do not have permissions to access. Run the
docker-compose
withsudo
- The files backing the mongodb are owned by a user you do not have permissions to access. Run the
Can I add my own dependencies?
Yes. Just add them to requirements.txt
.
Can I use a GPU acceleration?
Yes. See Docker's guide on leveraging GPU in docker containers.
The docker compose never returns!
You can send a SIGTERM to the program while it's in the foreground with ctrl-C or you can run the docker-compose
command with the option --abort-on-container-exit
Sensor Data Questions
What is the sample rate?
The sensor data is sampled at 60Hz. Each message contains a batch of 12 samples.
What is the noise level?
The noise level varies highly from testcase to testcase and from shelf to shelf due to environmental factors such as nearby vibrations and electrical noise.
What is the max weight?
The sensors are rated for 20kg per plate.
Do I need to account for sensor nonlinearity?
No. The nonlinearity error is orders of magnitude below the baseline noise from the environment.
Are the absolute weight values reliable?
No. The absolute weight measured by the sensors is not zeroed and may drift over long periods of time (hours or days). Relatively changes, however, are reliable.