Image Classification

Question 1

What is image classifcation?

Answer 1

Given an input image, assign image to one of a fixed set of categories

Question 2

What are the 6 challenges of image recognition?

Answer 2

The images could look different because of

1. Viewpoint changes (take from different angles, side, front, back)
2. Illumination (day, night, different lighting)
3. Deformation (shape of object may shape)
4. Occlusion (part of the object may be hidden)
5. Clutter (object may blend with background)
6. . Variation within class

Question 3

What is a straightforward example of data-driven approach towards image recognition?

Answer 3

kNN approach

Question 4

Why is kNN approach is not useful?

Answer 4

It is fast to train but slow in test. The direct opposite of an ideal algorithm

Question 5

How can you remove the noisy decision boundaries of kNN

Answer 5

By increasing the value of N is KNN.

Question 6

Why is there to split the dataset into train, validation and test?

Answer 6

Train - Model training
Validation - Tune the hyperparameter
Test - evaluate the model with tuned hyperparameter

Question 7

What is cross-fold validation?

Answer 7

Split the training set into multiple partitions and test and validate on each partitions.