MLS-C01 Question 80
Select 3You are working on an image classification problem and need to improve the performance of your deep learning model. Your dataset consists of 10,000 labeled images with significant variations in lighting and orientation. Which of the following steps should you take to prepare and preprocess your data to achieve better results?
- A
Apply data augmentation techniques such as random rotations, flips, and brightness adjustments.
- B
Normalize pixel values of the images to a range between 0 and 1.
- C
Resize all images to a consistent dimension required by the deep learning model.
- D
Remove all images with variations in lighting and orientation to ensure uniformity in the dataset.
- E
Scale the dataset by increasing the number of images using synthetic image generation.
Show answer and explanation
Correct answers: A, B, C
Explanation
Preparing data for a deep learning model involves applying preprocessing steps such as resizing images to a consistent input size, normalizing pixel values for consistent scaling, and using data augmentation to increase data diversity and robustness. Removing images with variations reduces dataset quality, and synthetic image generation is not a requirement in this context.
- A. Correct.
Correct. Data augmentation techniques like random rotations, flips, and brightness adjustments increase the diversity of the training data, making the model more robust to variations.
- B. Correct.
Correct. Normalizing pixel values to a range of 0 to 1 ensures consistent input scales for the model, which can improve convergence during training.
- C. Correct.
Correct. Resizing images to a consistent dimension is necessary for compatibility with deep learning models, which often require fixed input sizes.
- D. Incorrect.
Incorrect. Removing images with variations in lighting and orientation would reduce the diversity of the dataset, potentially leading to a model that performs poorly on real-world data.
- E. Incorrect.
Incorrect. While synthetic image generation can be useful in some scenarios, it is not necessary here as the problem specifically mentions using data augmentation techniques to handle variations.