Q1: Given a set of data points shown in figure 1, sketch (a schematic trend of change is enough, no need to calculate the exact numbers) a plot of how the average square distance of a data sample with respect to its corresponding cluster center will change as the number of clusters k increases.
Fig 1
Q2: Consider applying SLIC superpixelling to the image in Fig 2 (a). A sample result for 300 superpixels is shown in Fig 2 (b). Note how the resulting superpixels in uniform regions (e.g. those outlined in yellow) tend to be hexagonal. Why does this occur?
(a) (b) Fig 2
Q3: For Mean-Shift clustering, what happens to the number of clusters as the window size or bandwidth is increased? Explain your answer.
Coding:
• Apply skimage.segmentation.slic to 100075-Bear.png and show outputs through skimage.segmentation.mark_boudaries. Fix sigma to 5 for all experiments. o Compare the results for 50, 100, 200, and 300 segments and discuss. o Fix the segment number to 100. Compare and discuss the results when compactness is set to 1, 10, and 100. o Fix segment number to 100 and compactness to 10. Compare and discuss the results when the maximum number of iterations is set to 1, 2, 3, 4 and 5. • Apply sklearn.cluster.KMeans to the data in Q1. Change the initialization method to see its impact on final average square distance of all samples with respect to the clustering centers. Past Quiz Questions: • AY2122 quiz 2: Q1 (clustering)
