Currently each model expects that input data has been appropriately formatted beforehand. However, RGB images are a fairly common data format that many users will have for their data. It would be useful to support the submission of these images, rather than requiring users to reformat their data before using the tool.
What this would require is the ability to 1) specify that the image is an RGB image and 2) specify a mapping between indices in the RGB image and the necessary channels for the model. In theory, this could be expanded to include high dimensional images as well, with the user having the option to specify which indices match the required data for the model.
Currently each model expects that input data has been appropriately formatted beforehand. However, RGB images are a fairly common data format that many users will have for their data. It would be useful to support the submission of these images, rather than requiring users to reformat their data before using the tool.
What this would require is the ability to 1) specify that the image is an RGB image and 2) specify a mapping between indices in the RGB image and the necessary channels for the model. In theory, this could be expanded to include high dimensional images as well, with the user having the option to specify which indices match the required data for the model.