Divx5 LARGER than divx3?

Zarxrax · 7th March 2002, 17:55

Ok, we've all heard the reports of how divx5 is like 30% better and all, but I've done some tests here, and I think the results are astoinding! As I've mentioned in another thread, I've done some test encodes using anime, and the divx5 encodes only come out at similar quality as my sbc encodes. Well, here are some facts that support it!

I tested a 2 minute long clip from an anime with little motion. In divx5 I used constant quality to keep the drf at a constant level, and in nandub I set bitrate to 6000, freeze to 100, and locked the drfs to same levels that I tested in divx5.

DivX 5, no mpeg-4 options enabled:
DRF 2: 11.0MB
DRF 4: 6.17MB
DRF 16: 2.85MB

DivX 5, B-frames and GMC enabled:
DRF 2: 9.90MB
DRF 4: 5.68MB
DRF 16: 2.38MB

DivX 3
DRF 2: 9.65MB
DRF 4: 4.83MB
DRF 16: 1.63MB

Now, I have checked the divx encodes to see that they used the proper drf levels throughout, and indeed they did. Now what we see here, is DivX 5, Is much larger than Divx3 at the same DRFs! Only by enabling the new mpeg-4 features does it approach the filesize of the divx3 encode!

Now, as I mentioned, these tests were done on a clip with little motion. I also performed a small test on a 20 second clip with lots of motion, and in those tests the filesize of the divx3 turned out somewhere inbetween the 2 divx5 encodes.

I find this test results to be very interesting and shows that Divx5 may not compress better than Divx3 at all! I would love if some others could test this a bit more. I'd also be interested in comparisons to Divx 4 and Xvid as well.

Chibi Jasmin · 7th March 2002, 18:26

I have only done a few tests with quantizers all fixed to 2 (also in DivX3.11)....2 or 3 movies, some test clips...

DivX4 and DivX5 (without b, gmc, q) were always larger...

The only numbers I still have available are the following, so don't ask for more...

all without audio of course...

short Armageddon DVD clip

Quantizers all 2:

divx5+q+gmc+b 20570 KB
divx5+q+b 20770 KB
divx5+gmc+b 21800 KB
divx5+b 22036 KB
divx5+q+gmc 26862 KB
divx5+q 27346 KB
divx3 27836 KB
divx5+gmc 28760 KB
divx5 29352 KB

whole armageddon movie:

Quantizers all 2:

divx3 ca. 2,5 GB
divx5+q+gmc+b ca. 2 GB

Zarxrax · 7th March 2002, 19:01

Ok, so I got to thinking maybe Divx5 compresses better on live-action, and divx-3 compresses better on anime. So I've done more tests. This time I encoded 3000 frames from a very noisy video recorded on a camcorder. Once again, I'm still seeing the same results!

Divx5, no mpeg4 options:
drf2: 10.50mb
drf4: 6.05mb
drf16: 1.72mb

Divx5, b-frames and gmc:
drf2: 10.90mb
drf4: 5.53mb
drf16: 1.70mb

Divx3:
drf2: 8.23mb
drf4: 5.00mb
drf16: 1.46mb

Whats going on here?!

Zarxrax · 7th March 2002, 20:22

And finally, here is my final test for today. This test is not done terribly accurately, but it still shows the same results that I've been getting.
I encoded a full episode of an anime in divx5, at drf 2, with no mpeg-4 settings. I used the codecs built in crop and resize features. I used the smooth bicubic resize. The file resulted at 248MB.

I encoded the divx3 version in nandub, drf 2, and i used filters to crop and resize. I used nandubs presice bicubic filter, (which i assume is a neutral bicubic?). This is the only area where the test differed from the divx5 encode. I used soft bicubic in the divx5 and regular bicubic here. This file resulted at 244MB.

Finally, the divx3 and divx5 encodes come to about the same size. Had I been able to use a soft bicubic resize on the divx3, then I predict it would have resulted at about 230-240MB, a good bit smaller than the divx5. If I would have used b-frames and GMC on the divx5 clip, then theres a good chance that it would have come out a bit smaller than the divx3. However, these results still show that the divx5 codec simply doesnt compress as well as divx3!

Chibi Jasmin · 7th March 2002, 20:27

My numbers above all use the exact same AviSynth-input script (except Armageddon DivX3 full movie, which was with VDub Filters and Full Processing mode with anti-shit) and fast recompress (one time Nandub forcing DivX3 to use all quantizers 2, other time VDub 1.4.9 with DivX5 restricted to quantizers 2)

7th March 2002, 17:55	#1 \| Link
Zarxrax Registered User Join Date: Dec 2001 Posts: 1,219	Divx5 LARGER than divx3? Ok, we've all heard the reports of how divx5 is like 30% better and all, but I've done some tests here, and I think the results are astoinding! As I've mentioned in another thread, I've done some test encodes using anime, and the divx5 encodes only come out at similar quality as my sbc encodes. Well, here are some facts that support it! I tested a 2 minute long clip from an anime with little motion. In divx5 I used constant quality to keep the drf at a constant level, and in nandub I set bitrate to 6000, freeze to 100, and locked the drfs to same levels that I tested in divx5. DivX 5, no mpeg-4 options enabled: DRF 2: 11.0MB DRF 4: 6.17MB DRF 16: 2.85MB DivX 5, B-frames and GMC enabled: DRF 2: 9.90MB DRF 4: 5.68MB DRF 16: 2.38MB DivX 3 DRF 2: 9.65MB DRF 4: 4.83MB DRF 16: 1.63MB Now, I have checked the divx encodes to see that they used the proper drf levels throughout, and indeed they did. Now what we see here, is DivX 5, Is much larger than Divx3 at the same DRFs! Only by enabling the new mpeg-4 features does it approach the filesize of the divx3 encode! Now, as I mentioned, these tests were done on a clip with little motion. I also performed a small test on a 20 second clip with lots of motion, and in those tests the filesize of the divx3 turned out somewhere inbetween the 2 divx5 encodes. I find this test results to be very interesting and shows that Divx5 may not compress better than Divx3 at all! I would love if some others could test this a bit more. I'd also be interested in comparisons to Divx 4 and Xvid as well.

7th March 2002, 18:26	#2 \| Link
Chibi Jasmin Registered User Join Date: Oct 2001 Posts: 800	... I have only done a few tests with quantizers all fixed to 2 (also in DivX3.11)....2 or 3 movies, some test clips... DivX4 and DivX5 (without b, gmc, q) were always larger... The only numbers I still have available are the following, so don't ask for more... all without audio of course... short Armageddon DVD clip Quantizers all 2: divx5+q+gmc+b 20570 KB divx5+q+b 20770 KB divx5+gmc+b 21800 KB divx5+b 22036 KB divx5+q+gmc 26862 KB divx5+q 27346 KB divx3 27836 KB divx5+gmc 28760 KB divx5 29352 KB whole armageddon movie: Quantizers all 2: divx3 ca. 2,5 GB divx5+q+gmc+b ca. 2 GB

7th March 2002, 20:27	#5 \| Link
Chibi Jasmin Registered User Join Date: Oct 2001 Posts: 800	Just to add that... My numbers above all use the exact same AviSynth-input script (except Armageddon DivX3 full movie, which was with VDub Filters and Full Processing mode with anti-shit) and fast recompress (one time Nandub forcing DivX3 to use all quantizers 2, other time VDub 1.4.9 with DivX5 restricted to quantizers 2) Last edited by Chibi Jasmin; 7th March 2002 at 20:31.

7th March 2002, 19:01	#3 \| Link
Zarxrax Registered User Join Date: Dec 2001 Posts: 1,219	Ok, so I got to thinking maybe Divx5 compresses better on live-action, and divx-3 compresses better on anime. So I've done more tests. This time I encoded 3000 frames from a very noisy video recorded on a camcorder. Once again, I'm still seeing the same results! Divx5, no mpeg4 options: drf2: 10.50mb drf4: 6.05mb drf16: 1.72mb Divx5, b-frames and gmc: drf2: 10.90mb drf4: 5.53mb drf16: 1.70mb Divx3: drf2: 8.23mb drf4: 5.00mb drf16: 1.46mb Whats going on here?!

7th March 2002, 20:22	#4 \| Link
Zarxrax Registered User Join Date: Dec 2001 Posts: 1,219	And finally, here is my final test for today. This test is not done terribly accurately, but it still shows the same results that I've been getting. I encoded a full episode of an anime in divx5, at drf 2, with no mpeg-4 settings. I used the codecs built in crop and resize features. I used the smooth bicubic resize. The file resulted at 248MB. I encoded the divx3 version in nandub, drf 2, and i used filters to crop and resize. I used nandubs presice bicubic filter, (which i assume is a neutral bicubic?). This is the only area where the test differed from the divx5 encode. I used soft bicubic in the divx5 and regular bicubic here. This file resulted at 244MB. Finally, the divx3 and divx5 encodes come to about the same size. Had I been able to use a soft bicubic resize on the divx3, then I predict it would have resulted at about 230-240MB, a good bit smaller than the divx5. If I would have used b-frames and GMC on the divx5 clip, then theres a good chance that it would have come out a bit smaller than the divx3. However, these results still show that the divx5 codec simply doesnt compress as well as divx3!