Hello all,
I am working on a classification algorithm for pictures and the model I made has a perfect accuracy. When I use my test images in the notebook by first using a learn.load() with my ImageDataBunch being from folder, it will successfully identify all of the images (they are separate from what the model has actually seen).
However, when I choose to go with single_from_class to help prepare for deployment it will completely get most of the same exact images it got right, wrong. I am loading the model in with:
learn = create_cnn(data2, models.resnet34).load(‘test’)
where data2 = ImageDataBunch.single_from_classes(path, classes, tfms=get_transforms(), size=224).normalize(imagenet_stats)
Any ideas as to what may be going on or why this is happening? Was my model just not as accurate to begin with than it said?
Thank you,
Zach