Did you compute the diagonal run length for all 8x8 blocks per Y Cb and Cr?  Because the number of (code, runlength) pairs per each block per Y Cb Cr must be taken into account when computing compression ratio..  The example she gave in part F5 of the homework is only for a SINGLE Y block, you must do the same for all blocks of Y Cb and Cr if i understand it correctly.  Anyone care to verify this?

Don't forget she said to Pad the Cb Cr values so they can be divided into 8x8 blocks.

Once you subsample Cb and Cr is possible you will get dimensions that are not divisible by 8 even after padding the image as a whole.  So before you can divide Cb and Cr into 8x8 blocks you must pad them so their dimensions are divisble by 8.  You always use 8x8 blocks never 4x4 blocks

Well for example.  The ducky image is 250 x 273 and after padding the image the dimensions come out to be 280 x 280 so each dimension is divisible by 8.  When you convert the image to YCbCr, each component is 280 x 280, but when you subsample Cb and Cr you divide each dimension by 2 which results in Cb and Cr being 140 x 140 (280 / 2  x 280 / 2).  140 is not divisible by 8 so its impossible to divide the subsampled Cb / Cr components into 8 x 8 blocks unless you pad Cb and Cr so they become 144 x 144 each for the Ducky image.  Y will remain 280 x 280.

You lost me.  Lol...when i convert to Y Cb and Cr i create three separate matricies of values to hold the respective values for each component.  When you subsample Cb and Cr you iterate through them and take 2 x 2 groups of pixels and average them together into a new matrix that will have each dimension be half the size.  At this point you have to pad Cb and Cr before you can split them into 8 x 8 blocks and perform DCT each block.  In the reverse direction you combine the 8x8 blocks for Y Cb and Cr back to three separate matricies but before you can do super sampling on Cb and Cr you have to unpad the extra rows.  After super sampling Cb and Cr will have the same dimensions as Y.