AI service/model for tagging wildlife images

Joe Austin · March 05, 2025, 03:30:38 PM

I'm looking for an AI service and model that is good with tagging wildlife images (species, etc). I'd like to use Ollama. Are there any models that work best for this task?

Mario · March 05, 2025, 04:20:29 PM

All models available for Ollama are listed on their site: https://ollama.com/search

I have included all vision-enabled models in IMatch AutoTagger. When new vision-enabled models come out, I will add them if they are useful.

Note: You cannot use "any" model with Ollama. The model must have a specific format suitable for Ollama, and for that it mus usually be converted, which requires special software and expertise.

Google has just released their SpeciesNet model. Maybe we see support for it in Ollama at some point in time.

I would consider giving Mistral and OpenAI a try. These are big models and they may already be able to do what you need.

Stenis · March 05, 2025, 10:54:59 PM

Quote from: Joe Austin on March 05, 2025, 03:30:38 PMI'm looking for an AI service and model that is good with tagging wildlife images (species, etc). I'd like to use Ollama. Are there any models that work best for this task?

Despite I have been to quite a few parks both in Africa and Asia since the nineteen seventies nature photography is not really my biggest interest when it comes to photography so I have had big problems before I started to use Google Lens to identify species of both birds and lizzards for example.I'm very impressed by Google Lens both for identifying species and translate foreigh languages.

It is also very good at identifying architecture and landmarks.

So if animals was my priority I think I should test Google. I use OpenAI now but haven't tested it on my safari pictures yet but I will soon since I have been asked by some friends going to East Africa for the first time to show some of my pictures.

What really stunned me with Google Lens was the translation it did of old hebrew exerpts from the Dead Sea Scrolls from Qumran Caves. I just took a picture of some of these old scripts and in a second I was able to read the meaning of these more than 2000 year old texts in Swedish. Also very helpful when travelling in countries using other languages and character sets then the ones you might be used to.

Never underestimate Google. Who have more common search relevant data than Google to train their models on?

Jingo · March 06, 2025, 01:48:09 PM

I will +1 for OpenAI... recently switched my testing from Mistral to it and so far it is has correctly identified the 3 birds I've thrown at it. Of course, the images made it is easy to see the bird and its unique markings.. but I'm impressed it figured out: White-throated Sparrow, Mallard and Hooded Merganser.... I'll be testing it more on some of the more difficult birds like hawks and non-descript sparrows to see how it does. Not sure how it would handle lizards.. but for a few cents, you should be able to test out a few dozen photos.

Mario · March 06, 2025, 02:33:32 PM

Do you see a difference between the Mistral Pixtral 12b and the large Pixtral (more expensive) model?

Stenis · March 06, 2025, 04:50:25 PM

Quote from: Jingo on March 06, 2025, 01:48:09 PMI will +1 for OpenAI... recently switched my testing from Mistral to it and so far it is has correctly identified the 3 birds I've thrown at it. Of course, the images made it is easy to see the bird and its unique markings.. but I'm impressed it figured out: White-throated Sparrow, Mallard and Hooded Merganser.... I'll be testing it more on some of the more difficult birds like hawks and non-descript sparrows to see how it does. Not sure how it would handle lizards.. but for a few cents, you should be able to test out a few dozen photos.

Glad to hear since I started using OpenAI. Maybe I will find it good enough too.

Jingo · March 06, 2025, 05:58:48 PM

Quote from: Mario on March 06, 2025, 02:33:32 PMDo you see a difference between the Mistral Pixtral 12b and the large Pixtral (more expensive) model?

For me - I didn't notice a big difference but OpenAI is producing some great results "out of the box"!

Mario · March 06, 2025, 06:01:42 PM

Quote from: Jingo on March 06, 2025, 05:58:48 PM
Quote from: Mario on March 06, 2025, 02:33:32 PMDo you see a difference between the Mistral Pixtral 12b and the large Pixtral (more expensive) model?
For me - I didn't notice a big difference but OpenAI is producing some great results "out of the box"!

Very well. No model (yet/ever) does it all equally well. It all depends on which data it was trained on and how.
That's why IMatch is supporting multiple models and I may add more cloud-based AI's in the future, when I see advantages (Google, Antropic, ...)