Best Service to find Landmarks

[kiwilink]

I have been using the default Ollama in Autotagger to try and find out where I shot images.  I've also tried using Google and uploading the image just using the browser.

What is the best service I should try to match a Landmark? For example, I have some famous street shots in Quebec City, London, Tokyo, etc., but so far I'm not getting very many returns unless it's a famous spot like Big Ben.

Thanks!

Kiwilink

[Mario]

Detecting specific objects like landmarks or tourist spots usually requires a very large model, and of course a model trained with this kind of data.

The small models you can run on your PC in Ollama don't have this depth of information, unless it is a really prominent building like Big Ben or the Eiffel Tower.

Did you give Ollama some context, like "This photo was taken in Quebec City"? This may improve the results.

I've also tried using Google and uploading the image just using the browser.

Was this successful?

Did you try OpenAi and Mistral? They have huge models and a lot more depth in data than what you can run in Ollama.

Google may have trained their Gemini models with Google Maps data.
I did not include support for Gemini initially because I got much better results from OpenAI and Mistral (cheaper, too) when I've last testet Gemini. Same for Microsoft Copilot and Antropic.

The AutoTagger architecture I have developed is designed for change and flexibility.
Adding new AIs or new forms of interaction with AI is relatively easy to do. What is the best AI today may be something else tomorrow. AI is changed really fast.

I expect we'll see more specialized models or models with "mixture of experts" in the future. Maybe a model trained on Google Maps data, specialized to detect landmarks and places. With their street view data, Google should be able to do this.

[kiwilink]

Thank you for the great response.  I will try each one of these.