The .* (element-by-element multiplication) operator can proceed in the case where:
- every dimension that is not 1 in either InputWeight or TV.P, is the same between the two
- every dimension that is 1 in either InputWeight or TV.P is permitted to be any value in the other, and the length 1 will be replicated internally as many times as needed to match the other value
Because trailing dimensions are considered to have length 1, this does permit, for example, a 1024 x 768 matrix to be .* with a 1024 x 768 x 3 matrix -- the 1024 x 768 is implicitly 1024 x 768 x 1 so the rule of "you can have a mismatch if the length is 1" works out.
You could .* between 1024 x 1 vector and a 1 x 768 vector -- the implied replication can take place in both variables as needed.
What you cannot do is, for example, 1024 x 768 .* 1280 x 768. The 1024 is not 1 and does not match the 1280 that is not 1, so such a multiplication would be an error.
One of the more common ways this error occurs is if an RGB image is input but the code assumes that it was a grayscale image, especially if the code does something along the lines of
[numrows, numcols] = size(TheImage);
when TheImage is RGB this would result in a misleading value being stored in numcols.
