Vision: final assessment

This covers the whole course. Take it after the last module.

Final quiz

1. What is img[2][5] in a list-of-rows image?
2. What does a histogram of an image count?
3. Which kernel leaves an image unchanged?
4. What does opening do to small specks in a mask?
5. What is the luma greyscale value of pure white (255, 255, 255)?
6. What is the output size of a 3 x 3 valid filter on a 10 x 12 image?
7. Why does a mean filter turn one white speck into a blotch?
8. Which gradient direction does a horizontal edge between a dark top and a bright bottom have?
9. What does Otsu's method try for each candidate threshold?
10. Why should flood fill mark a pixel when it is pushed, not when it is popped?
11. Which interpolation should be used when resizing a mask of class labels?
12. Why are images warped by backward mapping?
13. What NCC score does a window have that is an exact negative of the template pattern?
14. Two stacked 3 x 3 convolutions with stride 1. What is the receptive field of one output value?
15. Why do CNNs need far fewer parameters than dense layers on images?
16. Precision is 1.0 at every prediction, and recall rises from 0 to 0.5. What is the AP?
17. In greedy NMS with threshold 0.5, a box has IoU 0.5 with a kept box. Is it removed?
18. Why does an RGB image from OpenCV look blue-skinned in an RGB viewer?

Practical assessment: Build a vision pipeline from scratch

Without hints or libraries, write a plain-Python program that finds, measures and classifies the objects in small synthetic images, and scores itself against ground truth.

Final project

An object counter for a factory camera: Build a plain-Python pipeline that takes small greyscale images of parts on a conveyor belt, with uneven lighting and noise, and reports how many parts there are, where they are and whether any are too small.

How it is evaluated

To complete this course