Another thing you can do with Gemma 3 is to ask it to explain a chart, business graphic, math formula or whatever to you.
For users who manage more than images in IMatch, this might be very helpful - because it transforms the visual data in a chart to something that can be used for searching, filtering and more in IMatch.
For example, using this prompt
Explain this chart to me in a few sentences. If the data can be transformed into a table, also return the table. Use Markdown to format your results.
I got these results. Original image with chart, resulting description in Metadata Panel, looking at the description in a Markdown editor (view the image in a new tab for a larger view):
Image8.jpg
Interesting. Just to clarify/confirm, you had the chart image in IMatch, with a description that included the statistical breakdown? Or, you just had the chart in IMatch, and the description, statistical breakdown and Markdown output were all generated by Gemma 3?
QuoteOr, you just had the chart in IMatch, and the description, statistical breakdown and Markdown output were all generated by Gemma 3?
Correct. IMatch has only the image of the chart as shown on the top left.
The description, break-down and table is done by Gemma 3.
It shows what can be done, and that there also can be some
mistakes.
In the table, the value for Windows 7 is shown as 11.0% instead of 11.3%. I've used the "large image size" setting in AutoTagger, but sometimes it gets one of the numbers in the table wrong.
Gemini 2.0 Flash
Lite does not produce a table.
Gemini 2.0 Flash (the more expensive model) produces a table and has the numbers correct, in 5 out of 5 tests.
This is why the AI companies offer different models, which have been trained with different foci / center of attention. Some are better for math and reasoning, some are better for image analysis.
Modern approaches like DeepSeek combine multiple models into one - this is named "mix of experts".