These improvements are really good to see!
I’m assuming this is accounting for the refractive index of both the case and the water (e.g. you measured the FOV with the assembled camera in water)?
Since FOV is not commonly measured in water, and also often written as separate horizontal and vertical components, it could be useful if you provided some direct comparison numbers
Given you’re doing color correction/compensation, it might be clearer/better marketing to describe it as that, instead of as “AI white balance”.
Saying “white balance” makes it seem (to me at least) like you’re just shifting the hue of the whole image so that white things appear white, but for a photo like the green one in the left of your image that would normally come up as still quite washed-out (see example below), whereas your improved result tries to restore/put back some of the colours that would normally be seen with the naked eye (if there was no water).
As a point of interest, are ‘normal’ camera settings applied before or after the AI processing occurs? I’m assuming exposure has to be applied beforehand (since it directly influences the image capturing process), but I’m unsure on others like brightness/contrast/saturation/etc. (or perhaps they’re not available with this camera?)
In case anyone is interested, here’s a comparison image with manual white-balance/hue-shift correction, and the basic visibility improvement code I discuss in this comment: