NCA-GENM Question 145
Select 3You are part of a team developing a multimodal generative AI model that integrates text and image modalities. Under the supervision of a senior team member, you are tasked with writing a script to preprocess image data for training. Which of the following steps should you include in your script to ensure the data is prepared correctly?
- A
Normalize image pixel values to a range suitable for the model (e.g., 0 to 1).
- B
Extract features from text captions associated with the images using a pre-trained text encoder.
- C
Resize all images to a consistent resolution required by the model.
- D
Randomly shuffle the image dataset to avoid training biases.
- E
Manually annotate each image with bounding boxes to improve multimodal alignment.
Show answer and explanation
Correct answers: A, C, D
Explanation
When writing scripts under supervision for preprocessing image data in a multimodal generative AI project, it is critical to normalize pixel values, resize images to a consistent resolution, and shuffle the dataset to reduce training biases. Tasks like text feature extraction and manual annotation fall outside the scope of preprocessing image data in this scenario.
- A. Correct.
Correct: Normalizing image pixel values ensures the input data is scaled appropriately for the model to process effectively.
- B. Incorrect.
Incorrect: While text feature extraction is important, it is unrelated to preprocessing the image data specifically, which is the focus of this task.
- C. Correct.
Correct: Resizing images to a consistent resolution is essential for ensuring compatibility with the model's input layer dimensions.
- D. Correct.
Correct: Randomly shuffling the dataset helps reduce biases during training by ensuring samples are not fed in a specific order.
- E. Incorrect.
Incorrect: Manually annotating images with bounding boxes is unnecessary unless the task explicitly requires object detection or localization, which is not mentioned here.