NCA-GENM Question 146
Select 3You are working as a junior developer on a team building a multimodal generative AI model that processes both images and text. Under the supervision of a senior team member, you are tasked with writing a script to preprocess the input data. The script should normalize image pixel values and tokenize text inputs. Which of the following steps would you include in your script?
- A
Convert image pixel values to a range of 0 to 1.
- B
Tokenize text inputs using a pre-trained tokenizer provided by the senior team member.
- C
Resize all images to a fixed resolution required by the model.
- D
Train the tokenizer on the raw text dataset from scratch.
- E
Add noise to the images to augment the training data.
Show answer and explanation
Correct answers: A, B, C
Explanation
Preprocessing multimodal data for generative AI models involves tasks like normalizing image pixel values, resizing images, and tokenizing text inputs. These tasks ensure the input data is in a consistent format for the model. More complex tasks, such as training a tokenizer or performing data augmentation, are outside the scope of basic preprocessing and would typically be handled by more experienced team members.
- A. Correct.
Correct. Normalizing image pixel values to a range of 0 to 1 is a standard preprocessing step for images in machine learning workflows.
- B. Correct.
Correct. Using a pre-trained tokenizer is a common practice for text preprocessing, especially when working with transformers, and this aligns with the requirement to work under supervision.
- C. Correct.
Correct. Resizing images to a fixed resolution ensures that the input dimensions match the model's requirements.
- D. Incorrect.
Incorrect. Training a tokenizer from scratch is a complex task and is typically not assigned to a junior developer under supervision. Instead, pre-trained tokenizers are often used.
- E. Incorrect.
Incorrect. Adding noise to images is a data augmentation technique for training, not a preprocessing step for input data.