Image Feasibility
Definition:
The image must be both clear and usable.
- Clear: Legible enough to support the question. Text, numbers, and visual features required to answer must be readable without ambiguity from poor resolution, pixelation, or blurriness.
- Usable: Workable enough to support high-quality annotation. Only skip an image if it is truly unusable — force nothing, but don't skip an image that's merely difficult.
✅ Pass
- All details needed to answer are sharp and clearly legible.
- The annotator correctly skips an unworkable image, or validly creates an annotation because the image is complex and workable.
❌ Fail
- The image is too blurry, pixelated, or distorted to confidently read the text, numbers, or visual features required.
- The annotator forces an annotation on an image that should have been skipped.
Skipping is a last resort, not a shortcut. To skip, select Annotator_Skip in the top right corner of the task in SuperAnnotate. Use this deliberately — the task is removed from your queue immediately and cannot be recovered. If you've already started your Handshake AI task, you must abandon it using the exit button in its top right corner before you can skip. If you haven't started one, no extra step is needed.
VQA rewards images you can reason over: tables, charts, graphs, dashboards, and dense infographics. Before you skip, check whether the image supports at least one question with a single verifiable answer, and whether it's rich enough to support more than one.
What makes a VQA image usable
- Readable data: labels, axis values, table cells, and legends are legible at full size.
- More than one element to compare, so the answer comes from reasoning rather than reading one printed value.
- A relationship, trend, or computed value you can derive and verify.
- Subject matter you can reason about accurately.
When to skip a VQA image
- Low quality — blurry, pixelated, or text too small to read.
- Too simple / not workable — minimalist scenes, generic portraits, movie posters, or flyers with no table, chart, or graph to reason over.
- Specialized knowledge — the subject requires expertise you don't have, so you can't verify the answer. Clarity is not enough on its own.
- Foreign language — foreign-language text is fine as long as none of your prompts rely on it. If you can write strong prompts that meet all other criteria without reading or interpreting the foreign-language text, the image is usable. Skip it only if the image can't be interpreted without that text.
- Toxic content — anything offensive or harmful.
Feasibility carries into the question. A clear image still fails this criterion if the question you write can't be verified from it: the answer must be readable or derivable from the image, with a single correct value and an explicit answer format. If you can only get there by estimating a value the image doesn't actually show, the image wasn't feasible for that question.
Do not use answer variance or a tolerance range to make an unverifiable question seem acceptable. “Any value within 5% is acceptable” allows multiple final answers and is unacceptable, even alongside whole-percent rounding. The image and specified rounding must support one verifiable final answer.
Chartography exception: If the task category says Chartography, an estimated value read from gridlines may use a stated accepted range or tolerance, and the model's wrong answer must fall outside it. All other tasks keep this rule. See Announcements.
For worked references, see Usable vs Unusable.
Knowledge Check
A fellow sees the image above and wants to write a prompt asking for the February CAC value for Channel 1. What is the correct action?