We evaluated your HEVC version 1.0.0.44 and have the following findings. 1.QP does not seem to change within a given encoded picture. 2.CU as B is rarely used in B pictures. 3.CU as skip for relatively static area is also less frequently used. CUs for some of these areas are even intra-encoded in P and B pictures. 4.QpI, QpB and QpOffsetB in the configuration file have big impact on the quality of the encoded video even though their corresponding ranges specified min and max Qp is sufficient large. This might indicate that the quality of the encoded pictures can be improved if the QP estimation is more accurate rather than heavily dependent on user input. 5.If we give the encoder enough flexibility to choose QP as in your sample cfg file, the quality of the encoded video is inferior to your H.264 encoder where we restricted the QP in a relative smaller range. (We have to restrict QP in your H.264 encoder to a smaller range to achieve acceptable quality as it too suffers this problem, at least for some earlier versions.)
Before item 4 above is resolved, we find it is difficult to achieve (sufficiently) better quality even at the same bit rate as your H.264. I was wondering if some of the following could be useful for your rate control:
a) Allocate significantly more bits for I picture than P, and significantly more for P than B.
b) Do a complexity analysis for each coding tree;
c) then allocate target bits according to the complexity;
d) Calculate average complexity and estimate average Qp for the whole picture. Varying your CT QP according to the variation of the complexity of the current CT.
e) When coding the CT, compare the number of bits generated for the current CT and the current picture and GOP so far against the corresponding targeted to adjust QP for the next CT.
Herewith I attached some your H.265 and H.264 encoded as well as PSNR and H.265 cfg files.
Regards,
Dongning
