don't get me wrong, appreciate you provide the source, and I agree this might be the closest we could get atm in terms of the references.

But still, we could measure slightly better than this, based on the angle differences the J-16 is opened, the center of the picture in Y-axis is roughly aliasing withe the left intake of the J-16 in the middle, I don't want to measure exactly how many pixels that is but let's just say roughly 10% may occur, so the average error in width is roughly 1-2% + pixel error for a quick estimation from your measurement, and that's assuming you are using the J-20 in the middle...

To do a more precise estimation, we could roughly reverse out the height of the camera using the same idea above, and this is Step 1.

For Step 2, we need the original picture (if I'm not wrong this is a cut from a larger picture) with ISO and other information of the picture (I have DJI Mavic myself, the lens is definitely more a fisheye, so boundary stretch is big).

But assuming the X and Y axis has the same angel error (unlike, but...), and ignoring the lens stretching, for now, based on the rough length/width estimation, you might looking for a 2-4% or even 6% error in length... that's from 20.04 to 20.46 after the error compensation.

But of course, this is a very quick estimation here, and I'm not sure exactly how many % the error exactly is, just think it should be slightly more accurate the 20.88 you got from raw pixel measurement.

yes, but the stretching and angle error on two jets are still different, not exactly useful to say...