There is a strong need for enhanced and more advanced object recognition capabilities in Qumagie. The current recognition system is limited to broad, generic categories such as Art & Music, Food, Nature, and similar predefined labels.
To make the feature significantly more useful, object recognition should become far more customizable. Users should have the ability to define and add their own custom “things” or object categories based on their specific needs, rather than being restricted to the default generic classifications.
For example, a user may want Qumagie to recognize specific product types, company-related objects, personal collections, equipment, or any other custom category relevant to their use case. This level of personalization would greatly improve the practicality and value of the recognition feature, especially for professional or specialized environments.
In short, Qumagie would benefit from a more flexible, user-configurable object recognition system that goes beyond standard generic labels and allows truly personalized classification.
This is quite a bit different than the generic models already implemented. For every different sort of item, an AI model needs to be created. It’s relatively straightforward to create an AI model to recognize people or cats or dogs or cars, etc. It’s quite another to recognize items specific to you. AI models would need to be created for each item and implemented. Now it’s going from somewhat generic that the whole world can use to specific to you. Does that make sense? It make it quite a bit more complex…
Hi,thank you for the explanation. it makes perfect sense, and I really appreciate you taking the time to walk me through the complexity involved.
That said, this kind of feature would be extremely valuable for me and my business. Generic models covering people, cats, dogs, or cars are useful, but they don’t really address my specific needs. Being able to recognize items that are meaningful to my workflow, objects unique to my activity, would make a huge difference in how I use my QNAP on a daily basis.
I fully understand that training a custom AI model for each category of item adds a significant layer of complexity. However, I was wondering whether QNAP might consider releasing some kind of guide, documentation, or toolset that would allow users like me to develop and train custom recognition models autonomously, directly on theQNAP, locally, without relying on cloud services.
Even a step-by-step tutorial, a dedicated app, or an SDK-like framework would be a great starting point. The idea would be to give more advanced users the ability to define their own object categories and train models on their own hardware, keeping everything private and local.