AI-901 ยท Common AI workloads
Which vision task returns what?
Image classification, object detection, captioning, tagging, OCR, and generating images from text prompts with diffusion-style models, including multimodal image understanding.
Key points
- Image classification: one label (or a few) for the whole image.
- Tagging: list of visual features.
- Object detection: each object with a bounding box.
- Captioning: a sentence describing the image (dense captions: regions).
- OCR: text from the image.
- Image generation: new image from a text prompt; a diffusion-style model starts from noise, denoises step by step.
- Multimodal models: image plus text question in one prompt.
Source: https://learn.microsoft.com/en-us/azure/ai-services/computer-vision/overview-image-analysis
Practice questions
Which vision task returns the location of each item as bounding box coordinates?
- Image classification
- Object detection
- Captioning
- OCR
Show answer
Answer: B. Object detection
Object detection returns each object with a bounding box. Classification labels the whole image, captioning writes a sentence, and OCR returns text.
Which statements about vision tasks are correct?
- Captioning describes an image in a sentence
- OCR extracts printed or handwritten text from an image
- Object detection returns bounding boxes with each item
- Image classification returns coordinates for every object
- Image generation requires bounding boxes as input
Show answer
Answer: A. Captioning describes an image in a sentence, B. OCR extracts printed or handwritten text from an image, C. Object detection returns bounding boxes with each item
Captioning writes a sentence, OCR reads text from images, and object detection returns bounding boxes. Classification does not give pixel-level coordinates, and image generation needs no bounding-box labels at prompt time.
True or false: a text-to-image model works by looking up the closest existing photo in a database.
Show answer
Answer: False
False. A diffusion-style model synthesizes a new image, starting from random noise and refining it guided by the prompt.
Practice 8 questions on this topic in Bitzu
Short lessons, instant explanations and spaced review of what you get wrong. Free on iOS and Android.
