Herewith a first one. ZzyZx asks

I would like to ask what exactly are we predicting with our models , GPS coordinates or displacement vectors?

from the the data page it says were giving a set of images and a list of its corresponding displacements xy. So my second question is are we to apply some kind of math function to compute the the needed x,y vectors the model is to learn from??

or we are to train on the raw x,y and perform the needed math function on test predictions???

We are predicting displacement vectors. The GPS is not known. This should be obvious.

You can either predict the displacement vector directly or convert to pixels and project the displacement in pixels. The model you use will determine which will be better for you.

The formula in the starter notebook shows you how to convert the vector to pixels. Note that you need to find the center first. Here is my utility functions that convert from the vector to pixels and back.

# Get image center in image pixels

def img_center_loc ( img ):

return img.shape [ 1 ] // 2, img.shape [ 0 ] // 2

# Given an image and target coordinates, return the target location in image pixels

def img_target_loc ( img, tgtx, tgty ):

ctr = img_center_loc ( img )

x1 = ctr [ 0 ] - np.round ( tgtx / CONST_X * img.shape [ 1 ] )

y1 = ctr [ 1 ] + np.round ( tgty / CONST_Y * img.shape [ 0 ] )

return int ( x1 ), int ( y1 )

# Given an image and target coordinates, return the target location in image pixels

def img_target_loc_with_ctr ( img, tgtx, tgty, ctr ):

x1 = ctr [ 0 ] - np.round ( tgtx / CONST_X * img.shape [ 1 ] )

y1 = ctr [ 1 ] + np.round ( tgty / CONST_Y * img.shape [ 0 ] )

return int ( x1 ), int ( y1 )

# Given an image and image location in pixels, return the target coordinates

def img_loc_coords ( img, pixx, pixy ):

ctr = img_center_loc ( img )

x = CONST_X * float ( ctr [ 0 ] - pixx ) / float ( img.shape [ 1 ] )

y = CONST_Y * float ( pixy - ctr [ 1 ] ) / float ( img.shape [ 0 ] )

return x, y

# Given an image, the image center and some location in pixels, return the target coordinates

def img_loc_coords_with_ctr ( img, pixx, pixy, ctr ):

x = CONST_X * float ( ctr [ 0 ] - pixx ) / float ( img.shape [ 1 ] )

y = CONST_Y * float ( pixy - ctr [ 1 ] ) / float ( img.shape [ 0 ] )

return x, y

The forward is given in the starter and the backward you should be able to work out with pen and paper easily. You can easily make a mistake as well, so double check that and let me know if you have issues with it.

do we get the field center(x1,y1) by using the formular stated in the starter notebook? or it is considered as (0,0) as stated in the data page?

also was the point constant defined in the starter notebook just random or it is what is to used ?

To get the center you need the formula - see the code above. The point constants

CONST_X = 10.986328125 / 2

CONST_Y = 10.985731758 / 2

are used, *I assume*, to convert from pixels to meters and I think is some kind of an average that the organisers got from their own data. It is just a scaling issue so any good model should be able to handle it easily. I mean, if you predict meters or feet (or pixels), it is the same thing, just in different scales. There is of course some rounding at play, especially at the pixel level, and I suspect this will in fact pollute your model, and for purity in the extreme you should perhaps address it by adding or subtracting a 0.5 here or there, but I doubt at this stage that pursuing this is relevant or productive in any way.