Skills Journal Flashcards
What is the difference between Artificial Intelligence, Machine Learning, and Deep
Learning?
a. Machine Learning is a subset of artificial intelligence. It allows the machines to
learn and make predictions based on its experience, which means data. ML can learn without explicitly being programmed.
b. Deep Learning is a subset of Machine Learning. The DL uses a multi-layer neural
network or other artificial neural networks for computation. DL can extract patterns
from raw data using Artificial Neural Networks (ANN).
Why do we train an ANN?
We train ANN’s to solve an optimization problem.
Can you list the basic necessary equipment that is needed to train, deploy and run a
DL project using the D900 camera?
a. D900 Camera
b. GPU + laptop
c. Training licenses dongle
Can you list at least 4 Deep Learning tools on the D900 camera and what are the
intended applications?
a. ViDi Check -> Automotive/Food & Bev for Assembly verification
b. ViDi Read -> Automotive/Electronics for OCR applications
c. ViDi Detect -> Electronics/Med/Pharma for Defect detection
d. Green Classify -> Identification/Sorting of objects
What is a perceptron, and what is it used for?
The perceptron is the basic algorithm for supervised learning, it is used to make predictions
Algorithm - Inputs are multiplied by weights and a bias to get an output
What is ViDiDetect?
ViDiDetect is an example-based anomaly detector. ViDiDetect trains on good
images and identifies any irregularity as a defect.
List the components of the ViDi Editor for the ViDiDetect tool.
a. Filmstrip
b. Tools Parameters
c. Image Panel
d. Tool Control
e. View Panel
f. Database Overview
What is a mask, and how is it applied to your image?
a. A mask is used to remove certain features from the inspection while highlighting
other areas.
b. Right-click on the image, select Edit Mask, click Apply Border, click the Apply
button, Yes to apply to all views, and click the Close button.
List the steps to utilize the Red Analyze tool in the Supervised Mode.
a. Collect images of Good and Bad samples.
b. Train images with the ViDiDetect tool. Bad images with graphical labeling of the
defects are mandatory. Good images are a plus.
c. Validate model on images with and without defects
When labeling a defect, is it okay to go a little outside of the defect? How about a lot?
It is okay to label a little bit outside of the defect, the algorithm will detect the
difference. Labeling a lot outside of the defect may cause the algorithm to see the
background as a defect as well, so that is not okay.
How are the Results from the Red Analyze tool in Unsupervised mode different than
in Supervised mode?
They are very similar. Supervised mode adds recall, precision, and f-score to the
results.
What are the two modes for the Red Analyze tool? How are they different?
a. Unsupervised, requires good images, detects random defects
b. Supervised, requires images with defects and each defect present be labeled, only
detects trained defects.
What types of applications is the Blue Locate tool meant for?
a. Locating, counting, fixturing, segmenting, and identification/qualification
List the steps to create a Layout Model.
a. From the Tool Menu, select Edit Models.
b. In the Model Editor dialog, select Layout Model.
c. Click the Add button to configure a region of interest that will define where to
search for one or more features.
d. Configure the region of interest over the desired area of the image.
e. Select the region, then select the feature(s) in the Feature Matching selector
menu, and in the Feature Count field, set the number of instances for all of the
features within the region.
f. Repeat c through e for each of the regions that will compose the model.
What is Recall?
How likely the tool is to find the desired object. Recall is the ability of the neural
network to be able to find the desired feature where it actually exists. It is denoted
as the number of correctly marked features divided by the number of labeled
features
