I am a bit confused on the sub-sampling portion of Y Cb Cr.

• If we are going to be converting back to RGB than how can we eliminate 3/4 of the Cb/Cr ? 
   
• Wont we need them to return to RGB?
   

As I understand it in sub-sampling we are taking the average value of every 4 values for Cb/Cr

When we get to the DCT transform that relies on an 8x8 matrix
What do we do if the subsampled portion of Cb/Cr is not divisible by 8?
Should I be just keep the same number of Cb/Cr values setting multiple indexes to the same value?

Good question!

After color conversion, subsample Cb(Cr).

If it is not divisible by 8, pad with zeros again. Then you can divide them into 8x8 blocks.

The HW3 description was updated for clarification.

--Kang