Contributions from research to this task will include:
| Status | Subtype | Assigned | Task | ||
|---|---|---|---|---|---|
| Resolved | • Ramsey-WMF | T225964 [SDC] Build a depicts tag suggestion tool that is powered by machine vision platforms | |||
| Resolved | Miriam | T228440 Computer Vision Consultation from Research |
Examples of issues of Google Vision API, related to the fact that their models are probably trained on unevenly distributed data: https://docs.google.com/presentation/d/1qfD3q9Ij79_luAKXNdVvZoPFsUvFQKcvIGmIvWC19z0/edit?usp=sharing
Some of these results show lack of relevant tags for images that come from underrepresented areas.
One of the main issues is that some tags are simply not there for images that are less "common" - and my doubt is that we end up reflecting this unevenly distributed information in Commons, thus making images from certain parts of the world less discoverable.
However, this is a problem that we might end up having with many of the available photo tagging tools, I don't think this is Google Vision-specific.
With the CAT tool now released, I think we can consider this work complete for V1 of this project. We'll sync up again after we have some real-world usage data. Thanks!