Trying a multi-class classification problem using the fast.ai library and lesson1 + lesson2 code.
Was able to run the code using directly the code for lesson1 without any major issue.
But I cannot understand how to interpret the probability values put onto images by the library.
For eg: in dogs vs cats lesson, most correct cats would be have probability 0, and for dogs it would be 1.
How do these values of probability transfer to the case of multi-label classification.
@parth_hehe did you find the answer to this? Struggling with the same thing myself. Currently trying to manipulate the ImageClassifierData.from_csv( ) to accept multiple labels as it seems it doesn’t right now.
From what I gather, if each image has multiple labels, one simply needs two directories: training and testing. Using the function ImageClassifierData.from_csv( ) that @gcmcalister mentioned, you have to put in the path to both the trianing and test directories, along with a csv file containing the paths to all training files. The funciton then uses this to make training and validation sets.
For an example of multi-label classification see this fast ai lesson: