JwumaEXIF and image provenance, the detail that decides whether a dataset holds up
· 2 min read · provenance, specification
EXIF records when an image was captured, on what device and often where. It is the primary evidence that a dataset is original field capture rather than scraped material. Most collection pipelines strip it silently, so it has to be preserved deliberately from capture through to delivery.
The short version
- EXIF is written by the camera at capture time.
- Gallery uploads, messaging apps and resizing routinely strip or rewrite it.
- In-app capture is the only reliable way to preserve it end to end.
- Location fields may need redaction while capture time is retained.
- Stripped EXIF cannot be reconstructed after the fact.
Scoping something like this?
We would rather talk about the problem than quote against a volume figure.
See capabilitiesWhat does EXIF actually prove?
That the image was produced by a camera at a particular time on a particular device, rather than downloaded. It does not prove the subject matter is what the brief asked for, but it does cleanly separate original capture from scraped files, and that is the distinction most often challenged.
For a buyer facing a rights question, this is the difference between having an answer and having a supplier's word.
Where does it get destroyed?
Almost everywhere, silently. Sending an image through a messaging app strips it. Uploading from a phone gallery often rewrites it. Server-side resizing and format conversion drop it unless explicitly configured not to. Some annotation tools re-encode images on import.
Each of these is individually reasonable and collectively fatal. The metadata is gone by delivery and no downstream process can restore it.
What about privacy?
EXIF frequently contains precise location, which you may not want and in some jurisdictions should not retain. The correct handling is selective: strip or coarsen location fields while preserving capture time and device.
Blanket stripping to solve the privacy problem destroys the provenance at the same time. Specify field-level handling rather than an all-or-nothing rule.
- Preserve capture timestamp and device
- Coarsen or strip precise location where required
- Prohibit re-encoding in the delivery pipeline
- Verify metadata on delivered samples, not source samples
People also ask
What is EXIF data?
Metadata a camera writes into an image file at capture, including timestamp, device and often location. It is the main technical evidence that an image is original capture.
Can EXIF be restored if it was stripped?
No. Once removed it cannot be reconstructed, which is why it has to be preserved deliberately from the moment of capture rather than addressed at delivery.
Should location data be kept in training images?
Often not. The right approach is field-level: coarsen or remove precise location while preserving capture time and device, so privacy is handled without destroying provenance.