Top-K accuracy in Computer Vision - Cheat Sheet & Quick Revision

Recall & Review

beginner

What is Top-K accuracy in machine learning?

Top-K accuracy measures if the correct answer is among the model's top K guesses. For example, Top-3 accuracy checks if the true label is in the top 3 predicted labels.

beginner

Why is Top-K accuracy useful in computer vision tasks?

Because some images can be ambiguous or have multiple possible labels, Top-K accuracy gives a more forgiving measure by checking if the correct label is within the top guesses, not just the top one.

intermediate

How do you calculate Top-5 accuracy for a classification model?

For each image, check if the true label is in the model's 5 highest probability predictions. The Top-5 accuracy is the percentage of images where this is true.

intermediate

Code snippet: What does this Python code do? preds = [[0.1, 0.7, 0.2], [0.3, 0.4, 0.3]] true_labels = [1, 0] # Calculate Top-2 accuracy

This code checks if the true label is in the top 2 predicted probabilities for each example. It counts how many times this is true and divides by total examples to get Top-2 accuracy.

beginner

What is the difference between Top-1 accuracy and Top-K accuracy?

Top-1 accuracy checks if the model's single highest prediction matches the true label. Top-K accuracy checks if the true label is within the top K predictions, allowing more chances to be correct.

What does Top-3 accuracy measure?

A If the model predicts exactly 3 labels

B If the true label is the single highest prediction

C If the model's accuracy is above 3%

D If the true label is among the top 3 predicted labels

Top-3 accuracy checks if the correct label is within the model's top 3 guesses.

Why might Top-K accuracy be preferred over Top-1 accuracy?

A It is easier to calculate

B It always gives higher accuracy values

C It allows for some prediction uncertainty by considering multiple guesses

D It ignores the true label

Top-K accuracy is useful when multiple guesses are acceptable, reflecting model performance better in ambiguous cases.

In Top-5 accuracy, what does the number 5 represent?

A Number of top predictions checked for correctness

B Number of classes in the dataset

C Number of training epochs

D Number of samples tested

The 5 means the model's top 5 predicted labels are checked to see if the true label is among them.

If a model has 90% Top-1 accuracy and 98% Top-5 accuracy, what does this mean?

A The model is better at predicting exactly one label

B The model's true label is often in its top 5 guesses but less often the top guess

C The model is overfitting

D The model's accuracy is decreasing

Higher Top-5 accuracy means the true label is usually within the top 5 predictions, even if not the top one.

Which scenario best suits using Top-K accuracy?

A When multiple answers could be acceptable or similar

B When only one correct answer exists and must be exact

C When the dataset has only two classes

D When the model is unsupervised

Top-K accuracy is useful when multiple possible answers are acceptable or close in meaning.

Explain what Top-K accuracy is and why it is important in evaluating classification models.

Think about how checking multiple top guesses helps understand model performance better.

Describe how you would calculate Top-3 accuracy for a model's predictions on a test set.

Focus on comparing true labels with the model's top 3 predicted labels.