CNN CNN - Convolutional Neural Networks Data Augmentation Techniques Questions and Answers 2 — Questions and Answers
Question 1: What does the ColorJitter augmentation randomly alter in a training image?
- Image geometry
- Brightness, contrast, saturation, and hue (Correct answer)
- Image resolution
- Aspect ratio
Correct answer: Brightness, contrast, saturation, and hue
ColorJitter randomly perturbs brightness, contrast, saturation, and hue to make models robust to varying lighting and color conditions.
Question 2: What is RandAugment?
- A random weight initialization scheme
- An automated augmentation policy that uniformly samples from a set of augmentation operations (Correct answer)
- A random architecture search method
- A random dropout strategy
Correct answer: An automated augmentation policy that uniformly samples from a set of augmentation operations
RandAugment simplifies augmentation search by randomly applying N operations from a fixed set, each at a global magnitude M, reducing the search space significantly.
Question 3: What is Test-Time Augmentation (TTA) used for in CNN inference?
- Speeding up inference time
- Applying multiple augmentations to test images and averaging predictions to improve accuracy (Correct answer)
- Augmenting the test dataset permanently
- Reducing model parameters at test time
Correct answer: Applying multiple augmentations to test images and averaging predictions to improve accuracy
TTA applies several augmented versions of each test image, collects the model's predictions for each, and averages them to produce a more robust final prediction.
Question 4: What does AutoAugment do in the context of CNN training?
- Automatically selects the best optimizer
- Searches for an optimal augmentation policy using reinforcement learning (Correct answer)
- Automatically adjusts learning rate
- Generates new architectures automatically
Correct answer: Searches for an optimal augmentation policy using reinforcement learning
AutoAugment formulates the augmentation policy search as a reinforcement learning problem, where the controller learns which transformations and magnitudes maximize validation accuracy.
Question 5: What is the 'Random Erasing' augmentation technique?
- Erasing random layers from the network
- Randomly selecting a rectangle in the image and replacing it with random pixel values (Correct answer)
- Removing random images from the dataset
- Erasing random weight connections
Correct answer: Randomly selecting a rectangle in the image and replacing it with random pixel values
Random Erasing selects a random rectangular region in the training image and replaces it with random noise or a constant value, helping the model become robust to partial occlusion.
Question 6: What risk does overly aggressive data augmentation introduce during CNN training?
- It always improves accuracy
- It can create unrealistic training samples that hurt model performance (Correct answer)
- It speeds up convergence too much
- It reduces memory usage
Correct answer: It can create unrealistic training samples that hurt model performance
Extreme augmentations can generate semantically invalid images that mislead the network, causing it to learn incorrect feature representations and degrading accuracy.
CNN CNN - Convolutional Neural Networks Data Augmentation Techniques Questions and Answers 2
What does the ColorJitter augmentation randomly alter in a training image?