Quick Draw. Intro to Classification

Intro to Image Classification

How would a robot use image classification to autonomously get through a maze?

Quick, Draw! (quickdraw.withgoogle.com)

Quick, Draw!

You can open this webpage in a new tab.

Copy link

Open in new tab

What is similar? What is different? Click to move on

Open Ended

After finishing 6 drawings click on a successfully identified doodle.

How is your drawing different from the 3 options that the computer recognized?

How is your drawing the same as the match? (Compare/Contrast)

Why didn't the neural network identify my brain! Click to move on...

Open Ended

Review this picture or one of your own drawings that the computer (neural network) did NOT recognize.

What about your drawing was the same?

Why didn't Quick Draw identify the image?

Play Again! Quick, Draw! (quickdraw.withgoogle.com)

Quick, Draw!

You can open this webpage in a new tab.

Copy link

Open in new tab

Open Ended

After playing again...

Did you do anything differently?

Did the neural network correctly identify more of your drawings?

If yes, why do you think the number of identified drawings increased?

Partner up and review BIG Data. Move to the next slide for questions.

Open Ended

Click on an image to review all the drawings of the class data set.

What image (class) did you and your partner select?

If your group found an image in the class that has been identified as a good drawing, report it to Google AI for review.

What was it about the bad images that led your group to report the mistakes?

Poll

How important is it for a driverless vehicle to be able to accurately identify and classify images?

• Not important, bad labels and classifications are part of life.

• Sort of important, accurate data helps save time.

• Extremely important, identifying and classifying objects incorrectly can be costly and potentially hurt someone.

Fill in the Blanks

Drawing Connections:

Explain and Justify how a robot would use image classification to get through a maze.