Can't get rid of Macroblocks :(

[cd090580]

Hi,

I have some problems with Divx encoding.

I have some videos I recorded with a standalone DVD Recorder in Max quality (about 9 Mb/s).

Now, I want to convert these to Divx to play them on a portable video device.

The portable video has a native screen resolution of 320*240 although it can play videos up to full D1 (ability to connect it to a TV).

So,
I convert my video to a resolution of 384*288 24 fps in Divx at 800 kbps and 128 kbps for audio.

Even with a so high bitrate for these resolution, there are lot of macroblocks in some areas in the videos

Some divx with a high resolution and a so low bitrate don't have macroblocks.

Here is 2 screens of the macroblocks in 384*288 (final video) (in bmp format to avoid enhancing macroblocks with jpeg):





and here is my divx settings:



I tried to to the resize with Virtualdub or avisynth but I got the same result.
Encoding is done with VirtualDub.

Is there any filters or procedure that can improve compressibility of the video or enhance it's quality ?

[jagabo]

single pass CBR isn't going to get you the best picture quality. Use multipass VBR, or single pass constant quality.

Also bump the Speed/Quality setting up a notch or two.

[GMaq]

Hi,
First of all DivX is a great Codec but it doesn't really shine at such small resolutions, Secondly if the final file size doesn't matter you should use "1-pass Quality Based" and in the presets on the "Codec" tab set it for "Extreme Quality" or even "Insane Quality" DivX won't throw too much bitrate at it in such a small frame resolution and you may even find the file is smaller or equal size in the end. In my own experience I've found that x264 is the only Codec that gives real quality images in the "Portable" resolutions. XViD performs better than DivX at this size but x264 is still superior. DivX and XViD are both intended for larger frame resolutions and that's where they show their stuff best. Try a test clip with the 1-Pass Quality based at "Insane Quality" and see how you like it. Hope this helps

[cd090580]

Ok, but for the 1 pass quality based, what value do I need to put for the quantizer (to keep the same average size as 800 kbps)

[jagabo]

Ok, but for the 1 pass quality based, what value do I need to put for the quantizer (to keep the same average size as 800 kbps)

There is no way of knowing. The size comes out whatever size is required to give you the quality you ask for.

With 2-pass VBR encoding you get a known file size but you don't know the quality.

With constant quality encoding you get a known quality but you don't know the file size.

I almost always use single pass constant quantizer encoding with Q=3. That's higher quality than most stuff you find on the internet. At that setting there is a small amount of macroblocking but you don't notice it at normal playback speeds. At Q=2 there are virtually no macroblocks but files can get quite large.

By the way, using Divx at settings higher than "normal" quality is pretty pointless. You will get very little difference in quality or file size (a few percent) but it will take much longer to encode.

[GMaq]

[quote="jagabo"]

By the way, using Divx at settings higher than "normal" quality is pretty pointless. You will get very little difference in quality or file size (a few percent) but it will take much longer to encode.

jagabo,
I would be willing to agree with what you said to a point, perhaps "Insane Quality" is too slow for the benefit, But there is a huge quality difference between Normal and Extreme Quality especially on CQ based 1 pass, On my machine there is very little difference in encoding speed between "Normal" and "Extreme Quality" anyway and I would think any difference would be relative regardless of machine speed.

[jagabo]

But there is a huge quality difference between Normal and Extreme Quality especially on CQ based 1 pass

There is no difference in quality between Normal and Extreme when using constant quality encoding. The difference is the file size. I've run many tests and rarely is there more than a 5 percent difference in file size. You may consider that "huge" but I don't.

<edit>

I just ran a test with a 90 second MPEG file. I resized the 720x480 frame to 640x360 with AVISynth's LanczosResize, then compressed with VirtualDubMod (Fast Recompress) with Divx 6.2.5 at Balanced and Extreme (Constant Quality, Q=3, no B frames):

Balanced: 24 seconds, 22,552 KB
Extreme: 72 seconds, 21,328 KB

Both files look identical, even at 4x enlargement. The Extreme file is about 5.4 percent smaller but took 3 times longer to encode.

Of course, if you do a lot of filtering the Divx encoding becomes a smaller part of the overall conversion time. For example, the same test as above, but using VirtualDubMod's Lanczos resizing filter, and Full Processing mode:

Balanced: 67 seconds, 21,978 KB
Extreme: 108 seconds, 20,860 KB

Since the filtering (resizing) is now taking much longer, Extreme is only taking 60 percent longer to encode than Balanced. The difference in file size is around 5 percent.

For those wondering why resizing in VirtualDubMod is so much slower: VirtualDubMod's filters all work in RGB mode, whereas LanczosResize in AVIsynth can work in video's native YV12 colorspace. VirtualDubMod has to convert the YV12 video to RGB, resize, then convert back to YV12 for Divx compression.

I did these tests on a Athlon 64 X2 3800+ using the 2-CPU version of Divx 6.2.5.

[jagabo]

Oops, pressed the wrong button. The MPEG file I used above was the 90 second Lost in Space trailer.

[GMaq]

jagabo,
Obviously how much benefit there is between encoding modes is going to be dictated by the source to be transcoded, on a brightly lit largely static source you may see no visible difference and see 1% file size difference, on a darkly lit action sequence it may be quite noticeable (that's where all of these MPEG-4 Codecs are weak anyway) and you could see a 10% difference in file size. The so-called 5% is not huge in light of the gross file size, however that 5% is where the fine visual details (like macroblocking) are noticed, so if it's the difference between the final product being 90% "Good" versus 95% "Good", I'll invest the extra time and bitrate. The main point is being happy with the final result which is a matter of opinion no matter what the numbers say

[lordsmurf]

Why not XVID instead of DIVX?

[jagabo]

The so-called 5% is not huge in light of the gross file size, however that 5% is where the fine visual details (like macroblocking) are noticed

As far as I can tell, the Encoding Mode setting in Divx effects only the width of the motion search algorithm, like Xvid's Motion Precision settings. When using constant quality mode motion search has no effect on image detail or defects like DCT ringing and macroblocking. It's the Quantizer value that controls how much detail is retained or thrown away.

Motion search precision effects details/artifacts when using CBR or multipass VBR encoding because the more motion vectors the codec can find the lower the quantizer values it can use.

[GMaq]

jagabo,
At any rate hopefully the OP is better off because we seem to be in 90% agreement and quibbling about the 10% we differ on, (That reminds me of something....Oh yeah, The Human Race!!) You've got me hands down with the empirical data, You obviously know DivX inside and out! The point I was trying to make was more philosophical than mathematical dealing with opinion and perception of the final product, Anyway I've learned my "something new" for today!

[cd090580]

Ok Thanks I will try that even if it takes lot of time to encode.

[cd090580]

After several tries, I noticed that mpeg of 352*288 capped from digital source (not done by me) and recoded to 384*288 in Divx (by me) looks very good without blocks.

Maybe the problem is because my source was analog and digitalized by my standalone dvd recorder ? If yes, how can I improve my analog capture ?

[jagabo]

After several tries, I noticed that mpeg of 352*288 capped from digital source (not done by me) and recoded to 384*288 in Divx (by me) looks very good without blocks.

Maybe the problem is because my source was analog and digitalized by my standalone dvd recorder ? If yes, how can I improve my analog capture ?

All the noise in analog captures makes video much less compressible. You could try wome spacial and temporal noise filtering but you'll probably find that, to get enough noise reduction to make the video significantly more compressible, you'll have to filter so heavily the video will look even worse.