Segmentation Flashcards
What is class segementation?
Each class gets a unique label
What is instance segementation?
Each instance of a class gets a unique label
Name one use for image segmentation?
Foreground extraction
Name two image segmentation methods?
K means clustering and convo nets
Describe the K-Means algorithm
1) Initialize clusters
2) Assign all points to hte closest cluster
3) Move clusters to the mean of the points in the cluster
4) Repeat 2-4 until convergence
How can we determine the number of clusters?
Perform clusteringen for several K’s and use the K where intra-class variance flattens out
How can we use K-means clustering for image segmentation?
1) On the histogram of the pixel intensities
2) Colour similarity
3) position and colour similarity
4) Others…
Name som advantages/ disadvantages of K-Means clustering for segementation
+ Fast and easy to implement
- No semantics and supervised methods often perform better
What is semantic segmentation
Assign a label to each segment
Why don’t we usually just use a sliding window and a classification network for segmentation?
Usually to computational expensive
How can we use a classification convo net for segmentation?
Change the fully connected parts at the end to convo. This gives a heat map of class probabilities. Use tranposed convolayer to upsample this heat map.
What loss would we use for a segmentation convo net?
Use the acerage cross-entropy
