Prepare a clearly visible face photo and distinguish detection failures, loading issues and changes in animal results.
The purpose of this guide
These are input-checking suggestions, not a photography standard tested to improve WhatZoo accuracy by a measured amount. They guarantee neither a particular animal nor identical results. Landmark estimation and the app's rule comparison are separate stages.
Choosing a photo
- Keep the whole face in frame, with eyes, nose and mouth visible.
- Start with an eye-level, frontal shot to simplify comparisons.
- Check focus and avoid backlighting or shadows that make the face difficult to see.
- Use a relaxed neutral expression when reducing expression differences. Smiling can affect the mouth-width and mouth-corner inputs used by the code.
- Check reflections and obstructions around the eyes. Removing glasses is not mandatory; try changing the light or angle if reflections obscure them.
- Use one person for single analysis. The current screen reports a separate error when multiple faces are detected.
Detection versus loading problems
If a face is small, blurred or covered, try another photo. However, the current general error can also appear after model loading or processing failures. If a clear face repeatedly fails, check the connection, reload and wait for model readiness. Changing the photo alone is not a guaranteed fix.
When the animal changes
Different poses or expressions can change rule matches. Even identical inputs can lead to random selection among highest-score ties. A new frontal photo does not establish one true animal type.
Change one condition at a time and record photos and results to track what changed. That comparison alone cannot tell you which result is more accurate.
Read the accuracy limits and analysis guide, then try the analysis tool.