Convert fastai ImagePoints to regular image

Part 1 (2019)
#1

I am running inference with a regression model in fastai that takes in images and predicts several (x,y) coordinates:

x, y, _, _ = model.predict(image, return_x=True)

I would like to be able to plot x and y externally, i.e., I need them in regular image format.
For x, I just do torchvision.transforms.ToPILImage(mode=‘RGB’)(img.data).
However, I can’t seem to figure out how to convert the y (coordinate tensor).
I can’t use fastai’s image.show(y=y) in this case (which works fine when I’m in Jupyter Notebooks).

Thanks for any help!

#2

I was able to get the coordinates into the right format by calling: scale_flow(FlowField(y.size, y.data), to_unit=False).flow.flip(1) on the y coordinate tensor :+1: