For the images that we want to add padding, I just wanted to make sure that their native centers before any padding correspond to GPS coordinates correct?
this is just to make sure that our results won't be off by a pixel.
I am not sure I understand what you mean by "native centers" but I can say that all the posistions are based on GPS coordinates and we provide only displacement vectors between the original and corrected positions.
thanks for your response, my apologies if my question was not clear.
So what you mean by "original" here is the image center no matter if its shape is a perfect square or not, correct?
Let's say you have a sample Sentinel image with a rectangular shape of (41,40) - you have 41 pixels on x and 40 on y. The native y-coordinate of image center is y_cnt=19.5 [in a coordinate system starts with 0 at pixel center].
You can edge-pad this image in two ways so it becomes a perfect square (41,41). Depending on which side you put the pads, the new image center will shift by half-pixel in the direction you put the pads on. Put another way, the new y_cnt after padding becomes 20 (in the new coordinate after padding), or either becomes 20 or 19 (in the original coordinate system, depending on which side you pad).
What I wanted to make sure is - the real center of the image as dictated by GPS is the center of the image file no matter if it is given as a square or rectangle?
Although the center and groundtruth for each image is based on GPS coordinates not pixels, adding one row (or column) to the images will make the center slightly off by a half pixel like you discribed. However, I believe this is not a significant error and won't affect your solution.
I am not sure I understand what you mean by "native centers" but I can say that all the posistions are based on GPS coordinates and we provide only displacement vectors between the original and corrected positions.
thanks for your response, my apologies if my question was not clear.
So what you mean by "original" here is the image center no matter if its shape is a perfect square or not, correct?
Let's say you have a sample Sentinel image with a rectangular shape of (41,40) - you have 41 pixels on x and 40 on y. The native y-coordinate of image center is y_cnt=19.5 [in a coordinate system starts with 0 at pixel center].
You can edge-pad this image in two ways so it becomes a perfect square (41,41). Depending on which side you put the pads, the new image center will shift by half-pixel in the direction you put the pads on. Put another way, the new y_cnt after padding becomes 20 (in the new coordinate after padding), or either becomes 20 or 19 (in the original coordinate system, depending on which side you pad).
What I wanted to make sure is - the real center of the image as dictated by GPS is the center of the image file no matter if it is given as a square or rectangle?
thank you!
No worries.
Although the center and groundtruth for each image is based on GPS coordinates not pixels, adding one row (or column) to the images will make the center slightly off by a half pixel like you discribed. However, I believe this is not a significant error and won't affect your solution.