AI-900 Practice Question: Describe features of computer vision workloads on Azure
What is 'image generation quality' evaluation — how do you measure if a generated image is good?
⚠ Common exam trap
Test-takers frequently assume objective, simple metrics like resolution or object counts are sufficient, but Azure AI-900 expects understanding that quality evaluation requires both automated distribution-based metrics and human judgment.
Answer choices
Why each option matters
Answer the question above first, then reveal the full breakdown to understand why each option is right or wrong.
Correct answer & explanation
✓
Metrics like FID (image distribution similarity) and CLIP score (prompt adherence), plus human evaluation
Image generation quality is evaluated using a combination of automated metrics and human judgment. FID (Fréchet Inception Distance) measures how similar the distribution of generated images is to real images, while CLIP score assesses how well the image aligns with the given text prompt. Human evaluation is also critical to capture perceptual quality that automated metrics may miss, such as aesthetic appeal and contextual coherence.
Answer analysis
Option-by-option breakdown
For each option: why learners choose it and why it is or isn't the right answer here.
- ✗
Only image resolution and file size — higher resolution means better quality
Why it's wrong here
Assessing image generation solely via resolution and file size conflates technical properties with perceptual quality. A high-resolution image can still contain severe artefacts, object-level errors, or poor composition, while a low-resolution image might be highly coherent (or, conversely, a file size can be large due to noise without meaningfully better content). Quality evaluation must consider aesthetic appeal, semantic fidelity to the prompt, absence of distortions, and scene coherence—none of which are implied by pixel dimensions or storage footprint.
- ✓
Metrics like FID (image distribution similarity) and CLIP score (prompt adherence), plus human evaluation
Why this is correct
FID (Fréchet Inception Distance) quantifies how well the statistical distribution of generated images matches real images in the feature space of an Inception network, with lower values indicating greater realism. CLIP score measures the cosine similarity between the image embedding and the prompt text embedding, capturing semantic fidelity to the prompt. Because each metric captures a different axis—FID for realism, CLIP for prompt adherence—and neither captures aesthetics or subtle coherence, combining them with human mean opinion scores (MOS) provides the most comprehensive quality assessment.
- ✗
Simply asking the model what score it gives its own output
Why it's wrong here
Asking the generator to rate its own output is fundamentally unreliable because the model has no external reference for what 'good' looks like; its internal scoring is typically based on learned priors that may not align with human perceptual judgment. It cannot detect its own failure modes, such as duplicated limbs or semantic contradictions, and it may be miscalibrated or biased toward its own generation distribution. Independent, objective measurements—like FID for realism and CLIP for prompt alignment—plus human raters are needed to mitigate these blind spots.
- ✗
Counting the number of objects correctly included vs. missing from the prompt
Why it's wrong here
Counting how many objects in the prompt appear in the generated image exercises only a coarse sense of prompt adherence. It fails to assess whether those objects are realistically rendered, correctly positioned, or visually coherent with the scene, and it ignores aesthetic quality, artefact presence, and safety risks. A generated image may include all requested objects yet still be low quality due to garbled textures or unnatural composition, so this approach cannot serve as a standalone quality metric.
Go deeper
Related to this question
About these practice questions
Courseiva writes every AI-900 question from scratch — 985 in total, each with an explanation and a wrong-answer breakdown. None are copied from real exams or dumps. Learn why practice questions differ from exam dumps →
JA
Written by Johnson Ajibi, MSc IT Security
Senior Network & Security Engineer · founder of Courseiva
This AI-900 practice question is part of Courseiva's free Microsoft certification practice question bank. Courseiva provides original exam-style practice questions with explanations, topic-based practice, mock exams, readiness tracking, and study analytics to help learners prepare for the AI-900 exam.