SD vs down converted HD at DVinfo.net

Pete Cofrancesco · June 16th, 2009, 11:53 PM

Recently I've switched from shooting in SD to HDV then down converting after editing and burning a DVD. But I got tired of spending the extra time editing and down converting, so I decided to do some tests to see if there really is a difference.

My conclusion was no. Even though I wanted the down converted HD to be the winner, I honestly couldn't see a difference. I shot various subjects and switched back and forth, took screen captures (which I could post) and they looked virtually identical. I kept re doing the test with the same results.

I've read many attest to down converted HD to be better but has anyone really done a side by side test and can post evidence to back up the claim? To me I think there's psychological bias in favor of dc hd. I'm going back to shooting SD and only shoot HD when the client or I would like hd the footage for future use. But honestly I can't ever see myself ever going back and re-editing.

Chris Hewitt · June 17th, 2009, 04:03 AM

Pete, the only real benefit I see of shooting HD and going from there is that for whatever reason, you have an HD source should you need that in future.
If they're shots that you plan to wipe when done, then probably easier to shoot SD just for the drive space and CPU issues.
For the price of a tape which you probably won't re-use, I'd shoot HD everytime.

Javier Gallen · June 17th, 2009, 05:49 AM

That is not accurate. It's camera dependant, so there's some models than gives a huge difference shoting on HD and downconverting to SD.

And besides that, keep in mind that if the SD mode is "standard DV", you can get better color resolution by downconverting. Just like having a 4:2:2 SD camera.

Another example: The goal of the 4k Red One is not to deliver a pristine 4k footage, but a 2k.

Of course, you have to balance the benefits of having larger renders and perhaps a harder editing with your actual equipment just for a little bit more quality.

Pete Cofrancesco · June 17th, 2009, 08:33 AM

Btw, this is with the Sony Z1u. I also didn't see a difference between down converting from the camera and the editor. Here are some stills the only difference I can tell is, SD yields a brighter exposure. Notice the blown license plate.

http://i43.tinypic.com/rlgphy.jpg
vs
http://i40.tinypic.com/2jct4cz.jpg

Robert M Wright · June 17th, 2009, 01:20 PM

How were you downsizing the edited HDV to SD?

As far as I can tell, most NLEs do a rather poor job of downsizing, and I can imagine most NLEs are no better at downsizing than having a camera convert to DV, on the fly, during capture (downsizing with some NLEs might even be worse). The best way I know of to downsize HD footage, is to use VirtualDub or AviSynth with Lanczos3 resizing.

I just took an unedited frame from HDV footage (shot with an HV20 at 24p) and downsized to 720x480 images, with Edius and with VirtualDub/Lanczos3 (using no comprssion with either Edius or VDub). I was going to upload the resulting images (as lossless PNGs), but apparently I can't upload images here. The Edius downsized image is notably softer than the VirtualDub/Lanczos3 downsized image.

Robert M Wright · June 17th, 2009, 01:47 PM

Quote:

Originally Posted by Pete Cofrancesco

Btw, this is with the Sony Z1u. I also didn't see a difference between down converting from the camera and the editor. Here are some stills the only difference I can tell is, SD yields a brighter exposure. Notice the blown license plate.

http://i43.tinypic.com/rlgphy.jpg
vs
http://i40.tinypic.com/2jct4cz.jpg

I downloaded and blew up your images. There isn't a big difference, but the first image does look slightly better to my eyes (magnified 250% on my 22" monitor). On a big screen TV, the difference might be perceptible, if looking closely (but not likely noticed if viewing casually).

The differences in the images I get from Edius and VDub/Lanczos3 is considerably greater.

I've pretty much concluded that it doesn't make much difference, downsizing HDV via camera or downsizing in an typical NLE, but sure does make a noticeable difference using Lanczos3 resizing to downscale (would be unmistakeable on something like a 60" HDTV).

Pete Cofrancesco · June 17th, 2009, 02:57 PM

The down converting was done when it was imported from the camera. I did a test prior to this comparing down convert from the camera vs the editor (I used MpegStream application which does a better job then FCP) and I didn't see a difference in quality and thats when it occurred to me to test it against SD.

I use TinyPic.com to post throw away images like this. Its free and you don't need to be a member.

Greg Laves · June 17th, 2009, 08:48 PM

I am with Robert on this one. I think the first pic looks slighlty better. But it could be down to exposure but I don't think so. It seems to be slightly sharper to me.

Pete Cofrancesco · June 17th, 2009, 09:33 PM

The SD would look better without the over exposure. The gamut on pc makes the grain more visible. Here are another scene.

http://i44.tinypic.com/bhn147.jpg
http://i43.tinypic.com/atw2fb.jpg

Robert M Wright · June 18th, 2009, 10:17 AM

I'm confused about the source of these images. Can you tell us exactly how (the process) each were generated?

We've got three possibilities:

1 - Shot as DV and captured as DV
2 - Shot as HDV and captured as DV (camera converting on the fly)
3 - Shot as HDV, captured as HDV and converted to SD by an NLE (or other software, on a computer)

Those images are also 853x480, so I assume they have been re-sized from 720x480, and also compressed (via JPEG). That does alter the image, but as long as the methodology is the same, and of high enough quality, for both images being compared, should still make for close enough to an apple-apple comparison (less than very high quality, in resizing to 853x480 or in JPEG compression, could potentially narrow the differences between the images though).

Peter Manojlovic · June 18th, 2009, 10:25 AM

Proper downconverting is a tricky little devil..Especially for 1080i sources. And it's as slow as molasses. I'm coming to the realization that one should shoot for the intended output..

But as mentioned earlier, my HDV projects can be called up at anytime for proper output..
That's the only thing pushing me to tape in HDV....

To do it properly, you need to BOB the fields to double framerate, resize, separate fields, choose the correct fields, and weave them back...

I just finished downconverting my first wedding, and the picture on a 50" Plasma was absolutely fantastic...I was shocked.
Would it have looked the same as my SD shoot?? I don't know...

But it's a pain to get a proper resize done...Anybody claiming resizing can be done on the fly, is throwing away information, and at that point, it becomes a subjective issue....

DO NOT Lanczos resize interlaced frames, without separating fields...It gives a deteriorated picture.....

If anybody's interested in my Avisynth script, just give me a holler.....

BTW, if there's issues in colour, you'll need to give workflow information to get to the bottom of any issues.

Robert M Wright · June 18th, 2009, 10:35 AM

Virtual Dub has a switch for resizing interlaced footage, so you don't need to manually get into bobbing and weaving like a prize fighter. (I can't say I've really tested to see if it works correctly though.)

To do resizing with Lanczos on a regular basis, using AviSynth (with proper script) would be quicker and easier though.

Robert M Wright · June 18th, 2009, 10:46 AM

Actually, I've found that resizing 1080i HDV to SD, with Lanzcos, treating the source as if it was progressive, isn't a bad way to deinterlace it to 30p while downsizing (kill two birds with one stone). It retains sharpness in static areas (extremely well), and usually blends motion areas pretty nicely. The beauty of it is there's no need for an algorithm to determine which areas of any given frame need to be deinterlaced (and thus, no mistaken deinterlacing occurs).

Peter Manojlovic · June 18th, 2009, 12:42 PM

Robert, please load your .AVI, go to the status flydown, and tell me what kind of FPS you're getting back...
I don't quite understand....Treating interlaced footage as progressive needs some sort of deinterlacing....
I'm not familiar with your method....I'll post some screenshots via my method when i get back....

Robert M Wright · June 18th, 2009, 01:15 PM

I just did a quick FPS test, with 1080i HDV (previously re-wrapped from an m2t container to an mpg container), using VirtualDub Mod (a version of VirtualDub that has been modified to be able to load MPEG-2 mpg files). Using Lanczos resizing (checking the box for resizing as interlaced content), it converted (to 720x480i60) at a little over 12fps, outputting as an AVI file using Canopus HQ as the output codec (the output codec will impact throughput considerably - H.264, for example, would obvously be much slower, but that has nothing to do with the resizing). This is with a Phenom 9500 (low end quad CPU).

If you don't check the box for resizing as interlaced content, you wind up with a result that has effectively been deinterlaced. Resizing 1080 lines to 480 lines that way, effectively blends the fields, rather than resizing them separately. The result is as sharp as you can get in areas where the image is static, and blends quite nicely in areas where there is motion. This works nicely for converting 1080i60 to 720p30 also. With AviSynth, you could effectively use similar methodology to convert 1080i60 to 720p60 (or 480p60).

June 16th, 2009, 11:53 PM	#1
Pete Cofrancesco Inner Circle Join Date: May 2004 Location: USA Posts: 3,005	SD vs down converted HD Recently I've switched from shooting in SD to HDV then down converting after editing and burning a DVD. But I got tired of spending the extra time editing and down converting, so I decided to do some tests to see if there really is a difference. My conclusion was no. Even though I wanted the down converted HD to be the winner, I honestly couldn't see a difference. I shot various subjects and switched back and forth, took screen captures (which I could post) and they looked virtually identical. I kept re doing the test with the same results. I've read many attest to down converted HD to be better but has anyone really done a side by side test and can post evidence to back up the claim? To me I think there's psychological bias in favor of dc hd. I'm going back to shooting SD and only shoot HD when the client or I would like hd the footage for future use. But honestly I can't ever see myself ever going back and re-editing.

June 17th, 2009, 09:33 PM	#9
Pete Cofrancesco Inner Circle Join Date: May 2004 Location: USA Posts: 3,005	The SD would look better without the over exposure. The gamut on pc makes the grain more visible. Here are another scene. http://i44.tinypic.com/bhn147.jpg http://i43.tinypic.com/atw2fb.jpg Last edited by Pete Cofrancesco; June 18th, 2009 at 01:20 PM.

June 17th, 2009, 04:03 AM	#2
Chris Hewitt Major Player Join Date: Apr 2006 Location: Broadstairs,Kent, England Posts: 225	Pete, the only real benefit I see of shooting HD and going from there is that for whatever reason, you have an HD source should you need that in future. If they're shots that you plan to wipe when done, then probably easier to shoot SD just for the drive space and CPU issues. For the price of a tape which you probably won't re-use, I'd shoot HD everytime.

June 17th, 2009, 05:49 AM	#3
Javier Gallen Regular Crew Join Date: Jul 2005 Location: Spain Posts: 67	That is not accurate. It's camera dependant, so there's some models than gives a huge difference shoting on HD and downconverting to SD. And besides that, keep in mind that if the SD mode is "standard DV", you can get better color resolution by downconverting. Just like having a 4:2:2 SD camera. Another example: The goal of the 4k Red One is not to deliver a pristine 4k footage, but a 2k. Of course, you have to balance the benefits of having larger renders and perhaps a harder editing with your actual equipment just for a little bit more quality.

June 17th, 2009, 08:33 AM	#4
Pete Cofrancesco Inner Circle Join Date: May 2004 Location: USA Posts: 3,005	Btw, this is with the Sony Z1u. I also didn't see a difference between down converting from the camera and the editor. Here are some stills the only difference I can tell is, SD yields a brighter exposure. Notice the blown license plate. http://i43.tinypic.com/rlgphy.jpg vs http://i40.tinypic.com/2jct4cz.jpg

June 17th, 2009, 01:20 PM	#5
Robert M Wright Inner Circle Join Date: Jan 2006 Location: Minnesota (USA) Posts: 2,171	How were you downsizing the edited HDV to SD? As far as I can tell, most NLEs do a rather poor job of downsizing, and I can imagine most NLEs are no better at downsizing than having a camera convert to DV, on the fly, during capture (downsizing with some NLEs might even be worse). The best way I know of to downsize HD footage, is to use VirtualDub or AviSynth with Lanczos3 resizing. I just took an unedited frame from HDV footage (shot with an HV20 at 24p) and downsized to 720x480 images, with Edius and with VirtualDub/Lanczos3 (using no comprssion with either Edius or VDub). I was going to upload the resulting images (as lossless PNGs), but apparently I can't upload images here. The Edius downsized image is notably softer than the VirtualDub/Lanczos3 downsized image.

June 17th, 2009, 02:57 PM	#7
Pete Cofrancesco Inner Circle Join Date: May 2004 Location: USA Posts: 3,005	The down converting was done when it was imported from the camera. I did a test prior to this comparing down convert from the camera vs the editor (I used MpegStream application which does a better job then FCP) and I didn't see a difference in quality and thats when it occurred to me to test it against SD. I use TinyPic.com to post throw away images like this. Its free and you don't need to be a member.

June 17th, 2009, 08:48 PM	#8
Greg Laves Major Player Join Date: Nov 2007 Location: Corpus Christi, TX Posts: 640	I am with Robert on this one. I think the first pic looks slighlty better. But it could be down to exposure but I don't think so. It seems to be slightly sharper to me.

June 18th, 2009, 10:17 AM	#10
Robert M Wright Inner Circle Join Date: Jan 2006 Location: Minnesota (USA) Posts: 2,171	I'm confused about the source of these images. Can you tell us exactly how (the process) each were generated? We've got three possibilities: 1 - Shot as DV and captured as DV 2 - Shot as HDV and captured as DV (camera converting on the fly) 3 - Shot as HDV, captured as HDV and converted to SD by an NLE (or other software, on a computer) Those images are also 853x480, so I assume they have been re-sized from 720x480, and also compressed (via JPEG). That does alter the image, but as long as the methodology is the same, and of high enough quality, for both images being compared, should still make for close enough to an apple-apple comparison (less than very high quality, in resizing to 853x480 or in JPEG compression, could potentially narrow the differences between the images though).

June 18th, 2009, 10:25 AM	#11
Peter Manojlovic Major Player Join Date: Jun 2007 Location: Hamilton Ontario Posts: 769	Proper downconverting is a tricky little devil..Especially for 1080i sources. And it's as slow as molasses. I'm coming to the realization that one should shoot for the intended output.. But as mentioned earlier, my HDV projects can be called up at anytime for proper output.. That's the only thing pushing me to tape in HDV.... To do it properly, you need to BOB the fields to double framerate, resize, separate fields, choose the correct fields, and weave them back... I just finished downconverting my first wedding, and the picture on a 50" Plasma was absolutely fantastic...I was shocked. Would it have looked the same as my SD shoot?? I don't know... But it's a pain to get a proper resize done...Anybody claiming resizing can be done on the fly, is throwing away information, and at that point, it becomes a subjective issue.... DO NOT Lanczos resize interlaced frames, without separating fields...It gives a deteriorated picture..... If anybody's interested in my Avisynth script, just give me a holler..... BTW, if there's issues in colour, you'll need to give workflow information to get to the bottom of any issues.

June 18th, 2009, 10:35 AM	#12
Robert M Wright Inner Circle Join Date: Jan 2006 Location: Minnesota (USA) Posts: 2,171	Virtual Dub has a switch for resizing interlaced footage, so you don't need to manually get into bobbing and weaving like a prize fighter. (I can't say I've really tested to see if it works correctly though.) To do resizing with Lanczos on a regular basis, using AviSynth (with proper script) would be quicker and easier though.

June 18th, 2009, 10:46 AM	#13
Robert M Wright Inner Circle Join Date: Jan 2006 Location: Minnesota (USA) Posts: 2,171	Actually, I've found that resizing 1080i HDV to SD, with Lanzcos, treating the source as if it was progressive, isn't a bad way to deinterlace it to 30p while downsizing (kill two birds with one stone). It retains sharpness in static areas (extremely well), and usually blends motion areas pretty nicely. The beauty of it is there's no need for an algorithm to determine which areas of any given frame need to be deinterlaced (and thus, no mistaken deinterlacing occurs).

June 18th, 2009, 12:42 PM	#14
Peter Manojlovic Major Player Join Date: Jun 2007 Location: Hamilton Ontario Posts: 769	Robert, please load your .AVI, go to the status flydown, and tell me what kind of FPS you're getting back... I don't quite understand....Treating interlaced footage as progressive needs some sort of deinterlacing.... I'm not familiar with your method....I'll post some screenshots via my method when i get back....

June 18th, 2009, 01:15 PM	#15
Robert M Wright Inner Circle Join Date: Jan 2006 Location: Minnesota (USA) Posts: 2,171	I just did a quick FPS test, with 1080i HDV (previously re-wrapped from an m2t container to an mpg container), using VirtualDub Mod (a version of VirtualDub that has been modified to be able to load MPEG-2 mpg files). Using Lanczos resizing (checking the box for resizing as interlaced content), it converted (to 720x480i60) at a little over 12fps, outputting as an AVI file using Canopus HQ as the output codec (the output codec will impact throughput considerably - H.264, for example, would obvously be much slower, but that has nothing to do with the resizing). This is with a Phenom 9500 (low end quad CPU). If you don't check the box for resizing as interlaced content, you wind up with a result that has effectively been deinterlaced. Resizing 1080 lines to 480 lines that way, effectively blends the fields, rather than resizing them separately. The result is as sharp as you can get in areas where the image is static, and blends quite nicely in areas where there is motion. This works nicely for converting 1080i60 to 720p30 also. With AviSynth, you could effectively use similar methodology to convert 1080i60 to 720p60 (or 480p60).