View Full Version : EX1 > SD better than shooting SD?


Rainer Mann
January 23rd, 2008, 06:08 AM
Hi
I have to shoot an interview for a media company (they will cut the stuff) They want to have 4:3/SD. I told them I wanted to use the EX which I will get this or next week. But they called it a "prosumer cam" - it would be better to shoot with a "big" SD cam. Well, as far as I know, EX is not "prosumer" at all and converting HD to SD should get a quality as fine as shooting directly on SD.
I wanted to shoot on EX with 4:3 grid in the viewfinder, opening a SD timeline in Final Cut Pro, taking the EX stuff to that timeline and exporting to an external Rekorder.
But this morning I read a thread about the bad quality resulting by that way... "steps" in lines etc--- recommendations? I will own an EX1 and still would have to rent an SD cam to get the best broadcast quality in SD 4:3 ? Hm... if so, ok - no problem. But I am really wondering...

Paul Kellett
January 23rd, 2008, 06:28 AM
I don't use Final Cut Pro,but i do use Vegas Pro 8 to work with my EX1 footage and there's nothing wrong with the footage once it's output to standard definition,on a DVD.The best way to get SD is to record HD then downconvert to SD.Hopefully some Final Cut users will reply for you but i can't see it being crap quality using Final Cut as a pose to Vegas.
Paul.

Rainer Mann
January 23rd, 2008, 06:45 AM
Thank you Paul. Hope to get replies about the quality of downconverting with Final Cut Pro and the best way to do so... I just don't "want" to believe having to rent an SD cam - but at least I want to deliver the best result to my customer !

Andrew Wilson
January 23rd, 2008, 09:16 AM
I think the footage looks really good once downconverted to SD but do some tests first. I find I get the slight stutter on a SD CRT television set since I shot in progressive and it's converted inside FCP to an interlaced format.

I haven't tried shooting in interlaced on the EX1.

Honestly, I think the EX1 really straddles the line between prosumer and professional leaning toward the pro side. It's still a sub-$10,000 camera and (I think) the best one available for the money today by far. It clearly delivers professional results from a prosumer looking package.

There's that old school expectation of a really big camera with Anton Bauer bricks stuck to the back and tape that records in 422 HD is what they need.

If you're brave, you could explain to your client that the EX1 shoots in 'industry accepted' XDCAM at full resolution and that since so few of your customers are still asking for SD, they're going to have to incur the expense of renting older equipment that they feel is superior. Plus, if they're cutting it, you should offer to get equipment that they natively support. I had a similar experience with a national broadcaster here. For them it was HDCAM or nothing. So I lost the job - but I'm also not supporting a $40,000 camera.

Jon Carlson
January 23rd, 2008, 09:38 AM
I had to use my EX1 on a re-shoot for a project that was originally shot in SD. It looked so much better than the original footage that I had to beat it up pretty bad to get it to match... besides scaling it down, I had to desaturate it and drop my brightness and contrast pretty heavily to get it to match.

Granted, I was matching it to footage shot on a JVC DV5000... not the greatest camera out there, but clearly the EX1 yielded superior results for an SD project.

I did have some minor problems scaling (Premiere Pro user) with artifacts. I took the footage into After Effects to scale and it looked fabulous.

FCP may or may not give them those same problems.

The only way I would see a big SD camera preforming better is with its bigger chips. I still don't know if it will be able to match the images that the EX1 is producing. Anyone have side-by-side comparisons available?

Rainer Mann
January 23rd, 2008, 10:30 AM
The stuff which I have to shoot will be cut together with other SD stuff. I am sure the EX stuff will look great, but will it match? Well, that's the question...

Alexander Ibrahim
January 23rd, 2008, 09:51 PM
At the end of the day use what the client asks for. It doesn't matter if they are wrong or whatever- just that they pay their bills. So- if they want you to shoot Digibeta or DVCPRO 50, rent the camera, charge them for the rental adding in a bit of markup get paid and get on with life.

That said they are dead wrong. The EX1 intercuts wonderfully with full size HD cameras. I've already intercut it with an F900 and a Varicam.

It also intercuts wonderfully with SD cameras. I've cut with it with XL-2 footage, and Varicam shooting DVCPRO 50. As mentioned elsewhere the only problem is that the footage is too sharp. That's easier to handle in post than the opposite problem of the footage being too poor in quality.

Also, do remember that the camera puts out 4:2:0 video. If they are planning on giving the footage an extensive post treatment (doesn't sound like it) then a 4:2:2 SD camera will match better- which leads you back to Digibeta or DVCPRO 50.

I see no problem with the HD to SD workflow within Final Cut Pro. I've done that with plenty of material, including footage from an HVX200. It is wonderful.

That said with the EX1 I prefer to use the camera's downscaling at ingest if I am working in SD.

Ali Husain
January 25th, 2008, 12:16 AM
lol. the ex1 gives you 960x540 4:4:4. you have to corrupt it with a decimation filter to get it to look like 4:2:2or god forbid 4:1:1. by downresing the 1080p image by half you also cut the noise by a significant amount--is it 6db? so you end up with the ultimate SD camera shooting widescreen 4:4:4, with the ability to shoot in poor light and still produce a low-noise image, probably comparable or better than a pd-170.

vegas does a beautiful job of downscaling video. however there was a thread somwehre here a few weeks ago that concluded you get *slightly* better downscaled image quality if you use after effects.

Ali Husain
January 25th, 2008, 12:20 AM
Also, do remember that the camera puts out 4:2:0 video. If they are planning on giving the footage an extensive post treatment (doesn't sound like it) then a 4:2:2 SD camera will match better- which leads you back to Digibeta or DVCPRO 50.


the ex1 can be corrupted to be 4:2:2

it is a 4:4:4 sd camera. for that matter the $650 hv20 is one too (except you'll get richer color and better lowlight for your extra $7000). :)

Alexander Ibrahim
January 25th, 2008, 12:29 AM
lol. the ex1 gives you 960x540 4:4:4. you have to corrupt it with a decimation filter to get it to look like 4:2:2or god forbid 4:1:1. by downresing the 1080p image by half you also cut the noise by a significant amount--is it 6db? so you end up with the ultimate SD camera shooting widescreen 4:4:4, with the ability to shoot in poor light and still produce a low-noise image, probably comparable or better than a pd-170.

Uh... That is all wrong. Very very wrong. You do the entire community a disservice by posting such egregiously wrong information.

Well except your core notion that the EX1 is a fine, if unwieldy, SD camera.

The EX1 is 1920x1080, and its 4:2:2 if you capture live over SDI, otherwise it records XDCAM to SxS which is 4:2:0

The Panasonic HVX200 is 960x540, but that is also only 4:2:2. It does output 4:2:2 DVCPRO HD onto P2 though.

You can certainly record 4:4:4 video using the EX1 as a source if you want, but the sensors are 4:2:2. There is no way to get 4:4:4 video off this camera.

In fact you have to spend quite a lot of money to get 4:4:4 sensors. I don't recall if RED 1 has 4:4:4, but your next choice for 4:4:4 video is the Sony F950, Arri D20 and Panavision Genesis

Alexander Ibrahim
January 25th, 2008, 12:31 AM
the ex1 can be corrupted to be 4:2:2

it is a 4:4:4 sd camera. for that matter the $650 hv20 is one too (except you'll get richer color and better lowlight for your extra $7000). :)

I already said it, but your information is so wrong I feel the need to refute it again.

The EX1 can NOT capture 4:4:4 video. Neither can the Canon HV20 or any other camera anywhere close to this price range.

David Heath
January 25th, 2008, 03:34 AM
Uh... That is all wrong. Very very wrong. You do the entire community a disservice by posting such egregiously wrong information.

The EX1 is 1920x1080, and its 4:2:2 if you capture live over SDI, otherwise it records XDCAM to SxS which is 4:2:0
I think you're overlooking part of Alis post, where he says it gives 4:4:4 *AT 960X540*. On the face of it, I'd agree with those figures - the EX will record to card 1920x1080 luminance samples, and 2x 960x540 chrominance samples per frame - 4:2:0. Downres to 1/4 of the luminance samples and you're left with 3 rasters of Y,U,V, each of 960x540 - 4:4:4 960x540, as Ali says.

Whether the practice works out as sweet as that would imply may be another matter for all sorts of reasons, and those simple maths take no account of interlace effects, for example, or alias effects being left over from the downconversion. But colour space numbers are not absolute measures of quality, rather chrominance/luminance ratios, and the above example shows why quoting them without reference to the luminance figure can be misleading.

As far as the original question posed, then if they are editing the material, one point in favour of using a dedicated 4:3 camera is the convienience - shoot, hand over tape or whatever, job done. Client goes away, no waiting for ingest, processing, output back out.

Ali Husain
January 25th, 2008, 06:54 AM
actually didnt' say anything about the sensors or about capturing at 4:4:4. if you downsample 1080p 4:2:0 by 2, the resulting image is 960x540 4:4:4, not 4:2:0. if you cut out an 852x480 portion of that image you've got 4:4:4 SD. if you want to get 4:2:2 SD from that, you have to make it worse by destroying chroma resolution in one direction.

Matt Davis
January 25th, 2008, 01:20 PM
But this morning I read a thread about the bad quality resulting by that way... "steps" in lines etc--- recommendations? I will own an EX1 and still would have to rent an SD cam to get the best broadcast quality in SD 4:3 ? Hm... if so, ok - no problem. But I am really wondering...

I've just come from a job cutting a project shot on 2/3" (a DSR-450 with $18k lens) & DVCAM, with a Z1 as B-Roll. We also happened to have a loaner EX-1 which I played with whilst waiting for the shooting crews.

Firstly, with the current software from Sony, everything zips into Final Cut quickly and easily. Astonishingly fast compared to my P2 experiences last year. Once detail was turned off and colour saturation raised a little, nice rich images.

I was interested in 720 at 50p, which I though would provide the benefit of longer shooting times and perhaps more success in becoming 50i when downsampled, yet easier to scale initially because of being natively progressive.

Just throwing 720p50 onto a DV PAL timeline and rendering it out provided results that didn't quite match the DSR-450 (that's an example of English Understatement). It wasn't horrible, just disappointing in comparison.

The image was lovely, luminsecent, radiant, rich in tonality. It just lacked the sparkle and zing of the original 720p footage. The zing and sparkle was in the DVCAM footage from the DSR450, and I would have said the EX1 matched it at native 720p (and you could buy two EX1s and more for just the DSR's lens).

I think the problem was in the scaling method, and possibly that I should be working with 1080 down to SD, not 720. It will require a resize from 1280x720 to 960x760 (distorting the pixels from square to CCIR-shaped) then an edge crop to 720x576 to get 4:3 PAL.

I will try custom settings in Episode, Compressor, QT Pro and Sorenson Squeeze. Just one little variance in a setting with compression software can mean the difference between art and compost.

My other alternative was to shoot 25Mbps quasi HDV and use an M15 deck to downconvert it to DV a-la-Z1, but taking the quality of the glass, Exmor chip and progressive nature with it. But I didn't get round to testing that, and now the EX1 is back with its rightful owner.

Alexander Ibrahim
January 25th, 2008, 01:45 PM
actually didnt' say anything about the sensors or about capturing at 4:4:4. if you downsample 1080p 4:2:0 by 2, the resulting image is 960x540 4:4:4, not 4:2:0. if you cut out an 852x480 portion of that image you've got 4:4:4 SD. if you want to get 4:2:2 SD from that, you have to make it worse by destroying chroma resolution in one direction.

Wrong. For different reasons than I originally thought, but still completely wrong.

You make one fatal assumption... that downsampling will throw out only those Y' (luma) samples for which there are no corresponding Cb and Cr samples.

Turns out you can't do that.

Y' is a calculated value, not a sampled one. The camera sensors sample R G and B. Then they gamma correct that to give R' G' and B'. These are given a weighted sum to lead us to Y'. There is no actual green data in a video signal in either rec 601 or rec 709.

So... if you did resample the image in the non-linear fashion you suggest you would seriously harm the Y' component. So, your reduced resolution image would not be 4:4:4 but rather 2:2:2.

Now, in other places on this forum I talk about the fact that the relative values of chroma subsampling matter a great deal for digital compositing.

However for visual perception the absolute value of the Y' sampling frequency is paramount to the sensation of "sharpness" in an image.

For some test images this method of chroma resampling you suggest would in fact work splendidly. The image wikipedia uses to demonstrate chroma subsampling is just about an ideal example.

http://upload.wikimedia.org/wikipedia/en/7/74/Chroma_subsampling_ratios.png

Also, solid fields of colors difficult to represent in 4:2:0 or 4:2:2 would benefit.

Most images however are quite irregular, and would suffer from having green "samples" misaligned. In other words it would be as if you'd applied a slur to the resampled image.

This is why video systems don't resample and rescale an image for lower resolutions, but instead simply downscale them. Its better to preserve the correct alignment of the green samples and the resulting Y' value than to try and create color information from thin air.

Perhaps more esoterically, if we were to follow your method we'd have to account for the fact that Kb and Kr have different values for SD and HD imagery. This would introduce errors in the conversion that would alter luma even further. The point being that not only will you throw away your resolution, but when you convert to SD from HD you are likely to distort colors, thus obviating the reason we started all these gyrations.

I suppose it is important to point out that a camera such as the EX1, could in fact sample 4:4:4 960x540 video with modest adjustments to its sensors and A/D conversion circuitry. The image would be biaxially pixel shifted and probably very pleasing. The moderating point is that these adjustments would have to take place in the camera before the image was recomposed as Y'CbCr. (Probably before it was converted to R'G'B'.)

Alexander Ibrahim
January 25th, 2008, 02:13 PM
Just throwing 720p50 onto a DV PAL timeline and rendering it out provided results that didn't quite match the DSR-450 (that's an example of English Understatement). It wasn't horrible, just disappointing in comparison.

The image was lovely, luminsecent, radiant, rich in tonality. It just lacked the sparkle and zing of the original 720p footage. The zing and sparkle was in the DVCAM footage from the DSR450, and I would have said the EX1 matched it at native 720p (and you could buy two EX1s and more for just the DSR's lens).

I think the problem was in the scaling method...

It sounds like there is a problem converting from rec 701 HD to the rec 601 DVCAM timeline.

You might want to check your sequence settings. You should make sure that FCP is not set to render everything as RGB... because that also causes shifts in tonality.

Here is the dialog I'm yammering about. You always want the "Render all YUV material in high-precision YUV" selected, but its a compromise between speed and quality. These days for SD sequences its fairly painless to choose the top two options- even on a MacBook

There may also be some proxy settings to adjust, but I don't think that's the case.

What I'd suggest is that you edit your HD footage on an HD timeline, then drop the edited sequence into your DVCAM sequence.

On other thing you might want to consider is using a 4:2:2 codec, like ProRes, instead of DVCAM for your SD sequence. There are some colors that won't ever translate from XDCAM to DVCAM. They may show up in a better quality codec.

Matt Davis
January 25th, 2008, 02:34 PM
Thanks Alexander - that's great info! I'm trying that right now.

On other thing you might want to consider is using a 4:2:2 codec, like ProRes, instead of DVCAM for your SD sequence. There are some colors that won't ever translate from XDCAM to DVCAM. They may show up in a better quality codec.

I do this with HDV too, and when doing projects with lots of Motion Graphics. The edit is done with native clips at DV, then the final render is to DVCPro50 for output to DigiBeta, or to Uncompressed for compression to whatever (assuming that all render files are ignored and we re-render to the final codec).

I've got a great path from HDV -->ProRez (used to be AIC) -->Compress --> Web Format, but like Rainer, I need a quick but high quality method to deliver DVCAM rushes to the client even if I shoot XDCAM-HD with an EX1, and those rushes had better be peers to DSR570s with broadcast glass, or else there's no point. EX-1 rushes to SD DVCAM (even if it requires a MacBook or MBP and a deck) makes the EX-1 into whatever the first 3-chip DV camcorder was - a game changer.

Alexander Ibrahim
January 25th, 2008, 05:26 PM
Thanks Alexander - that's great info! I'm trying that right now.

I am eager to hear the results

Ali Husain
January 26th, 2008, 01:12 AM
http://dvinfo.net/conf/showpost.php?p=776912&postcount=65
http://dvinfo.net/conf/showpost.php?p=753626&postcount=61

i won't be part of that. :)

rainer, before you worry or spend lots of money, you should shoot a well-exposed interview with any 1080p camera (hdv, xdcam, etc), downsample the footage using something reasonable (bicubic, bilinear, etc)--after effects, vegas, avisynth all do this--and compare that to the video you get from any native sd camera at any price.

good luck with your interview!

Matt Davis
January 26th, 2008, 05:01 PM
I am eager to hear the results

Right. Very unscientific (and I'm not referring to any notes here), but I tried

- to DV interlaced, default settings
- ditto, with Best and 10 bit High precision YUV
- to DV progressive (i.e. fields set to None, not Lower)
- the same, with Best and 10 bit High Precision YUV

I then tried adding a tiny smidgen of sharpening (FCP internal at 10). This was because I turned detail OFF on the camera. The person I borrowed it from had default settings, and I found the edges a little too zingy to my liking.

I was using a JVC 15" CRT monitor (not exactly Grade 1, but good enough) to view the FCP via FireWire to an M15 deck, then out to S-Video to the monitor.

The result for DV?

Setting to Best as per your description, ensuring that fields are set to None. I could also make a setting that would crop the 720p into a 4:3 hole.

If I get the chance, I'd like to do the same with 1080p as at 720p I think I got a small scaling artifact around edges just off horizontal that the sharpening drew attention to. I feel a bit guilty about saying this, but I think just the hint of sharpening seems to work too. Perhaps I should put Detail back on in the camera and repeat the tests.

So yes, I've now seen/achieved SD output that's streets ahead of the Z1 and almost equal to the DSR-450, just the DoF sets them apart. However, that's assuming I want 25p (which I do).

I tried putting 50p on a 50i timeline (so theoretically every frame should fit into a field), but it didn't seem to do that. I saw no field activity going on. But fields are so-o 1932 that 'am I bovvered?'