How to avoid compression artifact when shooting HDV? at DVinfo.net

Michael Maier · August 17th, 2005, 05:54 PM

Anybody cares to elaborate what would be the No's No's when trying to avoid compression artifacts while shooting HDV? Which shooting conditions would be most conductive for causing artifacts? Any special guidelines to avoid them?
Thanks.

Barry Green · August 18th, 2005, 01:28 AM

Most people seem to think it's related to panning or motion, and it's not. MPEG-2 handles that reasonably well. It has to do with the percentage of the screen that's changing, frame-to-frame.

Where you're going to see the footage break down is when everything in the frame changes, in an unpredictable way. Splashing water, or a rippling pool of water... fire will do it too. Zooming while panning can make it happen. Shooting a long telephoto shot, where there's heat waves rippling through the frame will do it. Backlit sparkly water like a lawn sprinkler can do it. Shooting shots where there's a lot of fine detail, like lots of blades of grass, can use up the bitstream, so an unexpected event in the middle of that can cause the issue -- a relatively static shot, for example, and then the sprinklers come on...

Where MPEG-2 shines is when there's little changing in the frame. So if you're shooting a beach shot of 80% sky and beach, locked-down, and the only rippling water is a little band in the middle, well, MPEG-2 can probably handle that. Change the shot to where there's nothing static, but you're shooting down at the waves and the screen's full of splashing waves, and it might choke horribly.

The more that changes from frame to frame, the harder it is for MPEG-2 to keep up with it. You'll notice on HDV demo shots from the manufacturers that they're usually locked-down shots; that's because in a locked-down frame there's a lot that doesn't change between frames. MPEG-2 encodes the changes between frames. Try filming a horse race like the Kentucky Derby, where there's horses charging hard around the track and the camera's wildly panning to keep up with them and there's a whole lot of action going on where their hooves are and there's a cloud of dust flying up -- MPEG-2 doesn't like that one bit. Horse racing is hard to watch on broadcast HD, it just totally falls apart and the whole legs/hooves area becomes a big blob of macroblocked ugliness surrounded by an apocalyptic swarm of mosquito noise.

Those are just some suggestions. Other areas where broadcast MPEG-2 falls apart are in things like the swimming and diving competitions at the Olympics, for example. Or a basketball game -- basketball is extremely artifact-y, although less so on ABC 720p than it was on CBS 1080i. Panning and zooming and a lot of action and a lot of detail = a severe challenge for MPEG-2 GOP-based compression.

Michael Maier · August 18th, 2005, 03:14 AM

Thanks Barry. It's good to know all that when considering the HD100 for a feature film with quite some action and a car chase.

Ash Greyson · August 18th, 2005, 01:18 PM

I have done a few HD shoots and the only problem I had was in a concert setting with lots of changing lights. It was a real disaster and most the footage was not even useable... I think Barry's assessment is right on...

ash =o)

Tony Behnam · August 19th, 2005, 10:45 AM

So what would you recommend when shooting fast scenes like a car chase?

Dominic Jones · August 20th, 2005, 11:31 AM

Another thing the Z1 (and presumably all HDV cameras, although I've not used anything else) absolutely hates is smoke - I did a noir-esque music video shoot recently on our Z1, with lots of backlit smoke, and again the large amount of randomly changing detail caused a very nasty compression on the smoky background.

I did another shoot with smoke after that, and this time used a smaller amount of well broken-up smoke to create a more even and less detailed bg, and this it handled much better...

Michael Maier · August 20th, 2005, 12:50 PM

Just remember that the Z1 is an interlaced camera. I think progressive would handle it better.

Barry Green · August 20th, 2005, 02:04 PM

Quote:

Originally Posted by Tony Behnam

So what would you recommend when shooting fast scenes like a car chase?

The smart-aleck response is "a format other than HDV".

If you're insisting on using HDV though, it may not do as poorly on a circumstance like that as you may think. The key is whether or not the motion is predictable. MPEG-2 can predict motion on objects and cope somewhat well if it finds the motion is predictable.

If you can keep your main subject reasonably stationary in the frame (basically a requirement of watchable cinematography anyway), that'll help the MPEG-2 compression out enormously. Then if the rest of the frame is behaving in a way that the compression can try to predict motion (i.e., a sideways pan where a trash can is in THIS position in this frame, and THIS position in the next frame; well, MPEG-2 can make a prediction that that block of data will be in THIS position in the next frame, etc...) That kind of motion it can try to cope with.

What it *hates* is unpredictable motion. Hates that. Rippling waves of water, smoke (thanks for that example), fire, heat waves, strobelights... those things it can't predict, and those can cause the compression to fall apart.

Steven White · August 20th, 2005, 02:26 PM

The other thing to avoid in the car chase is shooting with a high shutter speed. While some films have made use of strobe-like shutters for high action scenes to give a "gritty" feel - doing so on an HDV cam would push the bandwidth of the images very high, making it tough on the compressor.

I've shot a lot of in-car driving footage... most of it looks very good. So I don't imagine there will be a lot of difficulty shooting car chases, unless you're shooting straight out the side of a car, and trying to resolve rapidly changing senery.

Keep good composition, focus on the subjects, try not to have too much jitter. Considering that's what I'd suggest to anyone filming a car chase - I don't see how HDV would be much of a hindrance.

-Steve

Ken Hodson · August 28th, 2005, 01:29 PM

I think Barry is overstating things a bit. Unless the Sony cams bloated resolution is more of a hinerance then I have been lead to believe.
I have yet been able to stress the JVC HD-10 into showing any compression artifacts. Using 720p and 6GOP probably is why. So I can only imagine the HD100 would be even better. So no worries about car chases scenes and the like. Don't forget these cams also offer 480p60 which is great for these type of scenes, and give you lots of creative room in post. Amazing motion capture and slo-mo's.

Hans ter Lingen · August 29th, 2005, 10:13 AM

Buy the Panasonic AG-HVX200 and your problems are over.

Michael Maier · August 29th, 2005, 10:59 AM

Quote:

Originally Posted by Hans ter Lingen

Buy the Panasonic AG-HVX200 and your problems are over.

Yeah, and the other problems start.

Marty Baggen · August 29th, 2005, 11:30 AM

I would like to see a test by someone with an HDV camera.

Shoot a scene that causes artifacting.

Reshoot with a softening filter to see if artifacting is reduced.

I often use a Pro-Mist or similar filter to soften the image. It would seem that if you can reduce the harsh edges of a scene, the compressor will have a bit easier time. Filters can also raise blacklevels, thus reducing contrast.... also making compression an easier task.

Just anecdotal observations.... I have no data or hard numbers to prove my case, and I have only done this in SD.

August 17th, 2005, 05:54 PM	#1
Michael Maier Trustee Join Date: Apr 2005 Posts: 1,269	How to avoid compression artifact when shooting HDV? Anybody cares to elaborate what would be the No's No's when trying to avoid compression artifacts while shooting HDV? Which shooting conditions would be most conductive for causing artifacts? Any special guidelines to avoid them? Thanks.

August 28th, 2005, 01:29 PM	#10
Ken Hodson Trustee Join Date: Aug 2003 Location: Vancouver BC Canada Posts: 1,315	I think Barry is overstating things a bit. Unless the Sony cams bloated resolution is more of a hinerance then I have been lead to believe. I have yet been able to stress the JVC HD-10 into showing any compression artifacts. Using 720p and 6GOP probably is why. So I can only imagine the HD100 would be even better. So no worries about car chases scenes and the like. Don't forget these cams also offer 480p60 which is great for these type of scenes, and give you lots of creative room in post. Amazing motion capture and slo-mo's. __________________ Damnit Jim, I'm a film maker not a sysytems tech.

August 18th, 2005, 01:28 AM	#2
Barry Green Barry Wan Kenobi Join Date: Jul 2003 Location: North Carolina Posts: 3,863	Most people seem to think it's related to panning or motion, and it's not. MPEG-2 handles that reasonably well. It has to do with the percentage of the screen that's changing, frame-to-frame. Where you're going to see the footage break down is when everything in the frame changes, in an unpredictable way. Splashing water, or a rippling pool of water... fire will do it too. Zooming while panning can make it happen. Shooting a long telephoto shot, where there's heat waves rippling through the frame will do it. Backlit sparkly water like a lawn sprinkler can do it. Shooting shots where there's a lot of fine detail, like lots of blades of grass, can use up the bitstream, so an unexpected event in the middle of that can cause the issue -- a relatively static shot, for example, and then the sprinklers come on... Where MPEG-2 shines is when there's little changing in the frame. So if you're shooting a beach shot of 80% sky and beach, locked-down, and the only rippling water is a little band in the middle, well, MPEG-2 can probably handle that. Change the shot to where there's nothing static, but you're shooting down at the waves and the screen's full of splashing waves, and it might choke horribly. The more that changes from frame to frame, the harder it is for MPEG-2 to keep up with it. You'll notice on HDV demo shots from the manufacturers that they're usually locked-down shots; that's because in a locked-down frame there's a lot that doesn't change between frames. MPEG-2 encodes the changes between frames. Try filming a horse race like the Kentucky Derby, where there's horses charging hard around the track and the camera's wildly panning to keep up with them and there's a whole lot of action going on where their hooves are and there's a cloud of dust flying up -- MPEG-2 doesn't like that one bit. Horse racing is hard to watch on broadcast HD, it just totally falls apart and the whole legs/hooves area becomes a big blob of macroblocked ugliness surrounded by an apocalyptic swarm of mosquito noise. Those are just some suggestions. Other areas where broadcast MPEG-2 falls apart are in things like the swimming and diving competitions at the Olympics, for example. Or a basketball game -- basketball is extremely artifact-y, although less so on ABC 720p than it was on CBS 1080i. Panning and zooming and a lot of action and a lot of detail = a severe challenge for MPEG-2 GOP-based compression.

August 18th, 2005, 03:14 AM	#3
Michael Maier Trustee Join Date: Apr 2005 Posts: 1,269	Thanks Barry. It's good to know all that when considering the HD100 for a feature film with quite some action and a car chase.

August 18th, 2005, 01:18 PM	#4
Ash Greyson Trustee Join Date: May 2003 Location: Tulsa, OK Posts: 1,689	I have done a few HD shoots and the only problem I had was in a concert setting with lots of changing lights. It was a real disaster and most the footage was not even useable... I think Barry's assessment is right on... ash =o)

August 19th, 2005, 10:45 AM	#5
Tony Behnam Tourist Join Date: May 2004 Location: Geneva, Switzerland Posts: 3	So what would you recommend when shooting fast scenes like a car chase?

August 20th, 2005, 11:31 AM	#6
Dominic Jones Major Player Join Date: Mar 2005 Location: Brighton, England Posts: 225	Another thing the Z1 (and presumably all HDV cameras, although I've not used anything else) absolutely hates is smoke - I did a noir-esque music video shoot recently on our Z1, with lots of backlit smoke, and again the large amount of randomly changing detail caused a very nasty compression on the smoky background. I did another shoot with smoke after that, and this time used a smaller amount of well broken-up smoke to create a more even and less detailed bg, and this it handled much better...

August 20th, 2005, 12:50 PM	#7
Michael Maier Trustee Join Date: Apr 2005 Posts: 1,269	Just remember that the Z1 is an interlaced camera. I think progressive would handle it better.

August 20th, 2005, 02:26 PM	#9
Steven White Major Player Join Date: Nov 2004 Location: Canada Posts: 547	The other thing to avoid in the car chase is shooting with a high shutter speed. While some films have made use of strobe-like shutters for high action scenes to give a "gritty" feel - doing so on an HDV cam would push the bandwidth of the images very high, making it tough on the compressor. I've shot a lot of in-car driving footage... most of it looks very good. So I don't imagine there will be a lot of difficulty shooting car chases, unless you're shooting straight out the side of a car, and trying to resolve rapidly changing senery. Keep good composition, focus on the subjects, try not to have too much jitter. Considering that's what I'd suggest to anyone filming a car chase - I don't see how HDV would be much of a hindrance. -Steve

August 29th, 2005, 10:13 AM	#11
Hans ter Lingen Regular Crew Join Date: Jan 2005 Location: The Netherlands Posts: 55	Buy the Panasonic AG-HVX200 and your problems are over.

August 29th, 2005, 11:30 AM	#13
Marty Baggen Major Player Join Date: May 2005 Location: monroe, or Posts: 572	I would like to see a test by someone with an HDV camera. Shoot a scene that causes artifacting. Reshoot with a softening filter to see if artifacting is reduced. I often use a Pro-Mist or similar filter to soften the image. It would seem that if you can reduce the harsh edges of a scene, the compressor will have a bit easier time. Filters can also raise blacklevels, thus reducing contrast.... also making compression an easier task. Just anecdotal observations.... I have no data or hard numbers to prove my case, and I have only done this in SD.