WMV Encode Tradeoffs for "Classic" & "Art

fausseplanete wrote on 12/13/2008, 4:39 AM
Encoding a (corporate event) vid in WMV subject to client's bandwidth and filesize constraints led to settings negotiation! I ended up with 300 kbps for 320x240=76800 pixels themselves of 4:3 aspect ratio giving a 16:9 display at 25fps (UK) . The least I could get away with before serious blocking crept in, especially during fades and slow image zooms. I know the latter are advised against technically but the Art must prevail if at all possible (it was).

Now having seen a vid by someone else for the same client I see they have chosen to use square pixels (480x270=129600 pixels), the rest unchanged. That's about 1.7 times more pixels than me, into the same bandwidth.

So the question arises - at this "pushing the limits" bandwidth, which is best, overall: square (but more) pixels or fewer (but stretched) pixels? One obvious factor is motion - stretched hence fewer pixels leaves more bandwidth for motion - presumably why HDV went that route. But the tradeoff at HDV image sizes and bandwidths might not apply at mine (320x240 @ 300kbps), because at the end of the chain is the psycho-perception of the viewer.

So I re-rendered my footage to the square pixels format then loaded both formats in parallel Vegas tracks. Then went along timeline toggling the upper track on and off. Conclusions:

Immediate impression: square pixels was better, due to sharper text of titles. However...

Stretched/fewer pixels gave obviously and consistently better dissolves/fades to simple scenes and better (more details, less blocky, more professional-looking) slow zooms of photo images. Sometimes (not always) gave better fast motion (but who would notice it?). Static images of human life on the whole were cleaner.

Square/more pixels (at same bandwidth) : Hard items like text (vid, photo or generated) obviously and consistently sharper. Occasionally "life" scenes were also - varied.

So I think for "by the book" (low rate of change, cuts no dissolves etc.) WMV of this sort of size (320x240), fps (25-30) and fixed bandwidth (300kbps), square/more pixels is best, but for "flowing/arty" style (docu, performance etc) then stretched/fewer pixels is best.

Anyone care to share any further knowledge/experience on this matter?

Comments

fausseplanete wrote on 12/13/2008, 4:48 AM
I suppose another factor for the overall experience of the viewer is Consistency. If sharpness/softness varies then the defects will be the more staccato and hence obvious. And if initial image is especially sharp then it sets an expectation on the quality of the rest of the footage - causing disappointment if it is not met. So maybe care is needed to set a realistic expectation in the initial images. Another kind of tradeoff!
Chienworks wrote on 12/13/2008, 4:57 AM
One technical issue to take into account is that not all players understand PAR. Many, especially older ones, will display the video with square pixels no matter what the settings in the file say. People using those players would see your video displayed 4:3 with everything squished horizontally and stretched vertically. For this reason i always advocate square pixels for web viewing and other "lowball" distribution.

Consistency is a big concern. It is disturbing to have the image suddenly dissolve into fuzz. Two other tradeoffs you might consider are smaller frames and lower frame rates. Try 320x180, which is also 16:9 and not really that much smaller visually than 480x270, but uses only 44.4% as many pixels. That means each pixel or movement gets over twice as many bits! If the motion is generally slow then also consider using half frame rate. Maybe 12.5fps is pushing it slightly, but i use 14.985 a lot here in NTSC land with excellent results. That means each frame gets twice as many bits again. Combine both of these together and your result should compare to the original 480x270 25fps as if it had been encoded at 1350kbps.
farss wrote on 12/13/2008, 5:06 AM
One parameter not mentioned is time between keyframes. That seems to have quite an impact on bandwidth requirements, similar to GOP length with mpeg-2.

Bob.
fausseplanete wrote on 12/13/2008, 5:40 AM
Great stuff, thanks for those tips, wasn't aware that some players don't respond to aspect ratio. In the client's case it's an intranet hence controlled and only Microsoft's WMP is (currently) used. So I am lucky for the present but must bear it in mind.

I seem to remember Microsoft's own encoder being hackable/upgradeable to an advanced encoding set of options, but have never dared use it mainly because I wonder how robust (e.g. playable by different players) it would be and even in a controlled intranet whether it would be unrecognized without upgrading all the players.