Bounding box detection error (Gemini 2.0 (experimental) API)

[Soclatos]

Description of the bug:

I used the latest Gemini 2.0 (experimental) API for image recognition. My test image is a matrix of letters, but the results returned and the visualization do not match. The bounding boxes do not correctly correspond to the letter positions.
I used the same code as https://github.com/google-gemini/cookbook/blob/main/gemini-2/spatial_understanding.ipynb, and the result is shown in the image.
I also tried other visualization methods, modified the code myself, but it still didn't work.
I also used Google AI Studio's Starter Apps for testing and encountered the same issue.
So I'm not sure if there's an issue with the normalized coordinates returned by the API.

Actual vs expected behavior:

No response

Any other information you'd like to share?

No response