How to handle uint8/[0,255] format HSI?

[HaleySchuhl]

Start a discussion about anything you would like
So far plantcv has added support for a handful of different hyper/multispectral image formats (#972, #956, #781) and @rayn-alex brought up a formatting incompatibility that could make sense to handle automatically.

"It might make sense to introduce a cast to float, possibly already in read_data depending on the input data type. So the Spectral_data objects are always float32/0-1. The RAYN Vision System camera for example produces uint8/0-255 multispectral images and I need to convert to float32 and rescale them to 0-1 in my pcv workflows because otherwise the results do not make sense."

This sounds like a good argument to cast Spectral_data during read_data, or even inside _find_closest / other ideas ??

[rayn-alex]

I think it makes more sense in read_data because then all Spectral_data objects have the same array data type. Is there anything against that?

I guess the only argument against needless conversion is the floating-point error, but I guess it's negligible for these applications, right?

[rayn-alex]

Here is a sample image from the RVS Camera
sample_set.zip

[HaleySchuhl]

Thank you @rayn-alex this is super helpful. And my apologies for my slower response. I would agree that handling various datatypes as soon as possible makes the most sense, and we can evaluate if the decimal accuracy has an impact practically on results.

How do you currently convert your images into a "plantcv friendly datatype" ?

[rayn-alex]

No worries.

Just like this:

spectral_data = readimage(filename=img_file, mode='envi')
spectral_data.array_data = spectral_data.array_data.astype("float32")  # required for further calculations
if spectral_data.d_type == np.uint8:  # only convert if data seems to be uint8
    spectral_data.array_data = spectral_data.array_data / 255  # convert 0-255 (orig.) to 0-1 range