GOP length and filesizes (HCenc)

[Movie-Maker]

Hi,
isn't the whole idea about using more P and B frames in a GOP to reduce file size ?
I was recently fooling around with GOP lenghts and p/b frames and I noticed that even though I use a very long GOP with B-frames I dont get that much of a space improvement.

I encoded using HCenc a typical home video mini-dv clip. I just encoded the first 2000 frames. First I used a GOP length of 1, and 0 B-frames i.e. all frames are I-frames. Filesize = 70.3MB

The for the same 2000 frames, used a GOP length of "15,2" i.e. one I-frame every 15 P-frames and 2 B-frames between the P-frames. Filesize = 70.1MB

The filesize remained same for all practical purposes. I was expecting a good amount of reduction with the use of more p and b frames. Obviously I am not understanding something correctly. Can you anyone please help clarify ?
Thanks!

[jagabo]

File size = bitrate * running time.

The GOP size doesn't effect the file size. But the longer the GOP the lower the bitrate you can use and still get the same quality.

When using constant quality encoding GOP size effects file size. Switch to constant quantizer mode with different GOP sizes. You'll see that the output file gets smaller as the GOP size increases.

[Midzuki]

And if you really want to play with GOP-lengths and B-frames,
give a try to TMPGEnc Plus --- with it, you can go far beyond
36 frames per GOP or 3 B-frames for each P-frame.
Besides, "crappy" quantization matrices will help you to get even smaller file sizes.

\\\\\

[Movie-Maker]

Thanks for the replies.

I did some tests for filesize and bitrate using a CQ of 6, and without specifying a MAXBITRATE. I did this for GOP lenghts from 1 through 18.
I noticed that as you go from a GOP length of 1 through 4, the filesize and bitrate drops dramatically but after about a GOP length of 5-6 the filesize and bitrate really levels out. There is not much to gain in terms of filesize and bitrate by using a longer GOP than 5 or 6.

See the chart I plotted below. On the left in red is the filesize and on the right is the highest bitrate of a point in the source that I used as an indicator for bitrate.

Now when I specify a MAXBITRATE (9800), things are slightly different. The difference between GOP 1 and 4 is not as huge. Its subtle, but its there. I guess it depends on the content, how much of it really spikes beyond the max bitrate you would have to stay within.




{edit}: I forgot to mention this is with using just P-frames in the GOP. When I add 2 B-frames, the filesizes and bitrates are equivalent of adding 2 P-frames, which I found odd too. I was expecting a larger difference. I used the same quantization for all I, P & B frames.

Now how much discernable degradation of video quality between say GOP 4 v/s GOP 15, thats a subject for another thread.

[jagabo]

What that graph is telling you is the average size of I frames vs the average size of B/P frames. With a GOP of 1 you have all I frames. The average frame size is the average size of the I frames. As the GOP length increases the average frame size asymptotically approaches the average of the B/P frame sizes.