I have a 2D array representing a binary image.

I do: ndxs=np.argwhere(obj_img==0)

to get an array of indices into the 2D image array that I can then use to replace the 0s there with nan values via:

obj_img[ndxs]=np.nan

However, when I do this, I get an error similar to this, IndexError: index 22 is out of bounds for axis 0 with size 22

I tracked it down to the way the 2D array of ndxs is being interpreted in the obj_img[ndxs], but I don't understand why. If I instead do,

obj_img[ndxs[:,0],ndxs[:,1]]=np.nan

this works.

When I use ndxs as the index into obj_img, what is python doing? Why is obj_img[ndxs] not the same as obj_img[ndxs[:,0],ndxs[:,1]]?

CodePudding user response：

You can use numpy built-in Fancy indexing to replace all values when equals 0:

obj_img = obj_img.astype(float)
obj_img[obj_img==0] = np.nan

Note that you first have to convert it to float since NaN is a special floating point sentinel value