CS 451 Quiz 26
Photo OCR and machine learning pipelines
What is Photo OCR?
Scanning documents with cellphones
Reading text contained in photographs
Blurring license plates in streetview images
Enhancing the contrast or resolution of photos so that text is easier to read for humans
Some stages of a machine learning pipeline may not involve machine learning
In what sense is pedestrian detection easier than text detection?
Pedestrians have a fixed "aspect ratio"
Pedestrians have less variability
Pedestrians have a fixed height
In order to detect objects with varying scale (e.g., pedestrians at different distances) in an image using a sliding window detector, we
train different classifiers, one for each scale (window size)
train one classifier at the smallest scale, then downsize the window region from the image
train one classifier at the largest scale, then upsize the window region from the image
Why do we use a "stride" > 1 pixel for a sliding window detector?
For greater accuracy
For greater efficiency
To accommodate different scales
Which of the following stages of the Photo OCR pipeline could be solved with a neural net classifier? Check all that apply.
Which of the following is NOT a good way to obtain a larger training set?
Creating synthetic training examples from scratch
Adding "distortions" to existing training examples
"Crowd sourcing" new training examples
Duplicating existing training examples
Hand-labeling new training examples yourself
"Ceiling analysis" tests the potential performance gain of each stage in the pipeline by
simulating that (only) this stage worked perfectly
simulating that this stage and all earlier stages worked perfectly
simulating that this stage and all later stages worked perfectly
What are Andrew Ng's last words in the last video?
You should be proud of yourself
You learned a lot
Make the world a better place
Thank you very much for having been a student in this class
I ranked at least 20 movies on the movie ranking survey by noon today
This content is neither created nor endorsed by Google.
Terms of Service