01
Help reviewers start with an organized queue
Inspection teams may receive photos of packaging, returned products, or equipment conditions. StayCharted AMT can learn photo categories your reviewers have already assigned, helping organize incoming pictures for the next review step.
AI-powered visual inspection here means image classification for triage. It does not locate defects within a picture, measure their size, or determine whether equipment is safe to use.
| Illustrative photo | Team-assigned category |
|---|---|
| Carton with a crushed corner | Packaging damage |
| Product casing with a visible scratch | Surface damage |
| Inspection photo without an apparent issue | No visible issue |
02
Use your inspection team’s labels
Choose one clearly defined classification task. Agree on what each label means, and have qualified reviewers resolve ambiguous examples. Include normal conditions as well as the visible issue categories you want to distinguish.
Include variation in camera angle, lighting, background, and product type. A “no visible issue” label describes the picture only; it is not a certification that the item is undamaged or safe.
03
Choose for the defects that matter
Compare both Vision approaches on unseen inspection photos, including rare defects. On the cashew benchmark, the Dedicated AI Image Model found 26 of 40 defects, versus 16 for the AI Image Classifier Model. Overall accuracy alone hid this difference. The image-classification guide below explains how the approaches learn; your inspection team should choose using its own missed-defect results.
04
Prepare pictures your model can learn from
Upload a ZIP with one folder per category, or a spreadsheet of public HTTPS picture links and category labels. Use consistent labels and examples that represent the lighting, angles, backgrounds, and quality of future pictures.
StayCharted AMT checks for category mismatches, conflicting labels, near-duplicates, and unreadable links. Review those findings before training. AMT’s own processed copies carry no hidden image metadata; your uploaded file is kept as sent. This processing does not hide faces, personal information, or text visible in a picture.
Add the inspector’s note or the customer’s message as a text column; the model reads it with the photo, and a row with only a note is answered from the note.
05
Human-in-the-loop AI for inspection review
- New inspection photo
- StayCharted AMT category and confidence
- Your application creates a review item
- Reviewer confirms the finding
- Your team decides the next action
Use file results or API predictions to prepare review queues. Your application controls assignments and actions. Low-confidence or unfamiliar pictures need review, and high confidence does not remove the possibility of error.
Evaluate missed issues as well as incorrect flags on unseen photos. Do not use photo categories alone for safety approval, claim eligibility, or other consequential decisions.
06
Use your model in the field
Inspectors can check a photo from the field in the StayCharted AMT app for Android, with an access code and no AMT account.
07
Common questions about visual inspection
Can it find a hairline crack or a tiny defect? Fine details may be lost during image processing. Evaluate your actual photos and defect sizes; this workflow does not promise detection of small or hidden damage.
Does it draw a box around the damage? No. These models classify the picture into your categories; they do not provide defect localization or segmentation.
Can a dedicated image model improve the result? It is an option to evaluate when category differences are subtle. Compare measured results on the same set-aside photos rather than assuming it will perform better.
Can it approve an inspection automatically? The supported use here is categorization and triage. Qualified people and your existing inspection process determine the final outcome.