Leaderboard

Thanks to both of you for the advice. This should be enough for us to go on. Bobb Menk

I saw the facebook post about the new auto-tagging feature. As the full article says, tagging is a relentless chore and one that I'd love to avoid. But I know it's the only way I'm going to retrieve my pictures in the way I want. I have played with many AI-driven tagging applications and none of them has so far looked even vaguely useful. So I thought I'd take up Daminion's offer of a test drive of their take on this tricky subject. So I threw a few random images at the test site to evaluate the results. I'm sorry that the image is so small but there is a limit of 295 px wide, which is a shame. I did 9 images. The AI engine processed them with lightning speed. But how accurate were the results? More to the point how USEFUL to me are they? Because the most accurate AI tag is useless if it is not meaningful to me. Of the nine, five produced a tag that was accurate (check marks), although only one would have been my first choice Frequently, there were tags that unhelpful to downright weird. For example, the picture of a deer had a suggested tag of "Adaptation" Some had no useful tags at all and the picture of a cat was identified as a dog! If these tags were auto-suggested. I would have to spend a lot of time dismissing unwanted ones. Conclusion: Nice try but there's a way to go before AI tags would be even remotely useful. What's your experience? Paul

I, for one, can not wait for this implementation. It is after-all an optional feature, that you can activate/deactivate as a value added service. I am certain that the format and file size limitations (of the web preview) will be taken into consideration when the final integrations is pushed out to release. In the final release, I am sure that you will be able to set the accuracy threshold. I will experiment with a prediction threshold percentage (E.g. all tags above 80% probability) that I consider reasonably accurate. I will set up a new temporary catalog for imports so I can review and approve all newly assigned Google Vision tags. Once I am happy with the auto-discovered metadata I will move these assets into the working catalog. My tagging workflow will become one of revision and minor correction where the AI algorithm misidentifies. For those of us who have hundreds of thousands of digital assets to process yearly, this will save months in the tagging workflow. I DO NOT see this integration replacing human tagging, but rather to serve as a tag suggestion feature to increase productivity. This AI development is new and cutting edge in the DAM industry and I predict that as the technology matures the accuracy will too. You also have to consider that Google Vision uses a generically trained model. You can invest the time and effort to train your own specific model on the AI platform if you have a super specific domain.