|
|
iDS
NXT Ocean: A simplified Artificial Intelligence workflow experience
Back
to Newsletter
Go to iDS NXT Ocean Product Page
Go
to iDS Cameras
Go to High-resolution Colour Cameras
Go to High-resolution Monochrome Cameras
Go to GigE Cameras
iDS NXT
Ocean editorial two of four: How to label images for the IDS NXT
ocean AI workflow
|
Throughout this four part series,
we will take you through the steps to implement your own AI based
image processing with the all in-one-solution: iDS NXT Ocean. The
four basic steps to your AI solution are: grab,
label, train and run AI.
The iDS NXT Ocean is an all-in-one package that provides
you with everything you need to create your own AI inference camera.
It significantly simplifies the workflow related to training and
running neural networks.
In part one, we described how to acquire image
data from the iDS NXT cameras for labeling, training and model creation.
Please refer back to our May Newsletter editorial for detailed information.
Click
here to see our editorial on the "grab" element.
In this,
part two, we will cover the steps required to upload images to the
iDS NXT Lighthouse web-based AI training service and explain how
to apply labels and annotations to the images for classification
and object detection tasks.
In order to Label images the web application iDS NXT Lighthouse
is used. iDS NXT Lighthouse is an intuitive online platform that
makes it possible to label and classify objects in just a few steps.
iDS
NXT Lighthouse is a cloud-based software solution that enables users
to manage images for training of AI based image processing tasks
and the creation of individual neural networks.
|
|
The first step in this process is to open iDS NXT Lighthouse
in a web browser and then proceed to log-in. As seen below, the
NXT Lighthouse platform has a simple-to-follow interface.
The next step is to open the "data sets"
section in the top left hand corner of the screen. In this section
users can add new image data for training and can view any existing
data sets. To create a new data set, you click on the large "+"
icon. When here users can enter a text title for the data set, a
brief description of the data set as well as add a thumbnail image
from a PC's local drive.
Next, click on the "image data" tab where images that
have been acquired can be added to a data set. Users will then click
"add", then "add folders" and then we click
"add" again on the resulting dialogue. When here users
can navigate to the folder on
the PC that contains the images needed for the intended data set.
Users will then see a preview of the images. After you choose the
appropriate labeling, click upload to complete the process.
|
For object classification with the iDS NXT all that is required
of users is for data sets to be created and images to be uploaded
and then appropriately labeled. Similarly for
object detection users will need to add labels to specific regions
within the image. In some cases - you can use the same data sets
for object classification and object detection. In this editorial
we will give an example on how to label for detection. Labeling
for classification is even simpler.
The first step of labeling for object detection, is to create the
bounding box annotation which is used to draw around each of the
objects in the images. By clicking on the first image in the data
set you see the full image, as well as a new toolbar on the right
of the screen for annotations. It's here that users should now expand
the boxes section, choose a colour, give the box a name to reflect
the object class and click the "add" button. This box
is now available as a template for the object detection annotation
process. Now users can draw boxes around each instance of the detection
class in the image and when all of the objects in the image are
annotated click "save" and "next" to proceed
to the next image. If dealing with multiple detection classes
the NXT Lighthouse platform allows users to add subsequent boxes
with different colours and labels.
Once users have completed the labeling process iDS Lighthouse displays
statistics on the number of images in the data set and the number
of instances of an object. To conclude the process, the last tab
on the "display annotations" provides an overview of all
of the labels, box names and layouts associated with this data set
allowing users to edit as required.
And just like that, the process of labeling and making annotations
for classification or object detection is complete.
|
|
In the next editorial we will discuss creating projects and
learn how to train and build our neural networks. The
NXT Ocean package is designed specifically for tasks related
to object analysis and classifications for quality assurance.
This workflow maximises the efficiency of your application as
it makes it a lot simpler and more flexible. The iDS
NXT Ocean package can be purchased as individual components
or in a kit to simplify initial purchase. See below for more
information on this kit.
Click
here to read more:
|
What
the iDS NXT Ocean kit includes |
|
|
iDS Rio camera with
1.6 MP Sony colour sensor
|
iDS NXT Rio is a fully-fledged standard industrial
camera that can execute neural networks with hardware acceleration
giving inference times of a few milliseconds thanks to the integrated,
specially developed AI core ("deep ocean core"). Thanks
to features such as C-Mount, GigE network connection with RJ45 connector,
a serial RS232 interface and REST web interface, this model is ideally
suited for use in industrial environments.
With the Sony global shutter CMOS sensor the
camera is suitable for customers with high demands on image quality,
dynamic range and sensitivity. It provides a resolution of 1.6 MP
at 3.45 µm pixel size. The iDS NXT Rio GS29016 is available
in monochrome or colour.
|
|
6 Month license for
iDS NXT Lighthouse
|
The iDS NXT Lighthouse cloud software allows
those without expertise in the area of artificial intelligence or
camera programming to train an AI classifier with their own image
data. As a web application, the necessary means to create the neural
network are immediately available, meaning you wont have to initially
set up your own development environment.
Subsequently you can start on your own neural
network straight away. Image processing can take place entirely
or partially on the camera FPGA, this reduces bandwidth and computing
load.
|
|
The
iDS NXT Ocean kit also includes: |
High quality
lens
|
Gigabit Ethernet
Cable
|
Camera power supply
with necessary adapters
|
Camera tripod
adapter
|
|
Specifications
|
iDS NXT Rio
GS29016
|
Sensor Type
|
CMOS Colour
|
Shutter
|
Global Shutter
|
Sensor characteristic
|
Linear
|
Resolution
|
1456 x 1088 Pixel
|
Pixel Size
|
3.45 x 3.45 um
|
|
Need
a price or more application information? Please
email Adept Turnkey or call our offices
Adept Turnkey Pty Ltd is "The Machine
Vision and Imaging Specialists" and distributor of iDS products
in Australia and New Zealand. To find out more about any iDS product,
please call Adept Turnkey at Perth (08) 9242 5411 / Sydney (02) 9905
5551 / Melbourne (03) 9384 1775 or contact
us online. |
|