Aliasing Artifacts
 

Reference beta footage from Anton Nelson on Vimeo:

Canon 720P 50FPS BETA FOOTAGE on Vimeo

Canon 7D 1080P 24FPS (beta footage) on Vimeo

Original 1080 & 720 files available to download for registered Vimeo users.

On download of the 720 50fps video, look at the jaggies around the edges of the metal sculpture, compared with the 1080 version.

The 1080 version looks very similar to the aliasing artifacts on the 5DMK2 around the edge highlights.

Of course this is a beta model.

Full res grab attached:

Wow, I don't see the technical problems, I see wonderful images with very little artifacts. Moving water is pretty hard for AVCHD...

Xavier it's not the codec or general compression but the in camera scaling.

see pic below.

Hi James. I can see the antialiasing scaling artifacts, but to me they are not as nasty as the blocking artifacts at the bottom right of the picture, in the out of focus area. It's not obvious in the stillshot, you need to look at the moving video. If this footage is really direct from the camera it doesn't look very good at all for anyone who is looking for shallow depth of field with attractive bokeh.

Richard

At first I was going to put this off to sharpening artifacts, as it does seem that these clips are at least moderately sharpened (probably in-camera). then I checked out some new clips rob galbraith posted today (the shots of the viewing platform engulfed by waves)

Rob Galbraith DPI: Autofocus, video and more

his 1080 clip appears to be completely without sharpening...yet the aliasing in the 720 clip is even more pronounced than those mentioned in this thread...to my eye, they look more like upscaled 16:9 480p clips than native 720 clips..

Lets hope this is a pre-production issue.

Yeah it is there on 720P footage, I don't see much on 1080P, hopefully it is not there.

Agree the 720p from Rob looks terrible.

I have just downloaded Philips 720p video from Vimeo. Looks good, blascks look a little hard and obviously the full frame and shallow DOF mask any perceptual resolution. But in shot 01:39:28 you can see a few jaggies on Philip's off cuts in the sink.

I wonder when changing from 1080p to 720p does the camera use the same picture profiles with custom settings?

It seems Canon has left the known problems of 5DM2 unfixed in this model so I wonder whether 7D gonna pass BBC's test.
So far 5DM2 footage is not approved by BBC for serious TV production due to these issues:

http://thebrownings.name/WHP034/pdf/...on_5D_DSLR.pdf

This is not accurate. A broadcaster can ban a camera from their own production. But how can a broadcaster ban footage from any capture device? How can even guess from where the footage has been produced?

This camera produces high quality professional footage to any broadcaster in the world if you know what you're doing. As matter of fact, as it happens with any digital camera since dv.

And, the 5DMKII has been used as a second camera on at least one Hollywood production I know about, probably more. Networks only want good quality footage within broadcast tolerances (ie., exposure, etc.) on whatever formats meet their requirements.

Are we concluding that the 720p is scaled up and actually sampled smaller?

I have seen Philip's film many times now and don't see really anything I don't like. I know we see the post grading and recompression and I am curious to see his raw footage. I see slight blockiness in the out of focus regions, but doesn't seem any worse than HDV and the horrid chromatic aberration on my first HD100. I just don't want jagged lines!

I took the Galbraith footage, and converted it using Cineform's Neo Scene.

It looks like Cineform is changing the levels, as with the 5D footage, giving the footage more detail in highlights and in the shadows.

Second, there may be some improvement in both the 720 and 1080p material with respect to aliasing an moire. But I will let you judge for yourself. See the attached grabs.

Cineform is changing the levels as compared to what? From what I've seen, the older version of Quicktime clips the blacks. The newer version raises the gamma. Cineform, on my system decodes properly and sets the levels to 16-235. You can also get the correct levels by re-wrapping the MOV files in Quicktime to M4V (or is it MP4?) and applying the computer to studio filter in Vegas.

Take a look at the histograms and you should be able to see what's going on.

Chris,

In your 720 orig and cineform grabs, are those from exactly the same frame? The water streaks from the splash look significantly different.
If they are then the recompression is doing a lot of changes.

Ted

Ted:
I forgot and pulled that frame from the end of the clips.. Cineform actually retimes, so I think that is why I was off. I will recheck.

Chris,

Thanks for the reply. Its interesting to note the alias smoothing that cineform seem to apply based on your samples. I'm especially curious to know what it does to the streaks of water and the railings since being in different angles of the diagonal exhibit the stairstepping prominently. Can you post again exact same frames for better comparision?

It would be nice if you can find that same frame which you used for the 720 resolution, that moment when water splash from the rocks is at its longest.

Thanks for doing this tests.

Ted

I am posting two new sets just, in which I tried to be sure I was getting same frame.

I am not sure I am seeing that much difference in group.

Chris,

The 1080 samples are both ok to my eye, its easy to tell because theyre from the same frame. I think you may have moved a frame in the 720 samples. check again.

Ted

Chris,

I'm interested in your tests because,

1. I do my edits on a 720 24p timeline. easier on my hardware, and because i shoot a lot of 60p slomo stuff.

2. I also use a similar intermediate codec only mine is HQ. So i'm making comparisons.

Thanks in advance.

Ted

Of course if you pixel-peep you will find defects with any camcorder. Even on the 100.000$ one!!! If I shoot moving water with my HDV cams I get nothing but low-level youtube quality. Just the fact that the AVCHD of the 7D can cope with moving water and not break away is good news (at least to me!).

Canon should decide to either build the ultimate full-frame 1920x1080 CMOS (who said night-vision?) or to upgrade their scaling algorithms. The 7D has a couple of processors, maybe they can do better than read one line of every three...

Hello Xavier,

I wasnt much into pixel peeping but more into the codecs for editing. You see, HQ only runs on Edius. Its similar to Cineform but I abandoned it when prempro 2 came out and was to heavy for the processors at that time. Now that processors are faster I might see if Prem with cineform can do the task. Its integration with after effects is a plus.

But then if its just the same or only slight difference, I'll stick with edius and HQ.

Ted

We need to have 2000 thousands things under-control!

C'mon aren't u guys being a little harsh.
 

Guys are you serious. By inspecting every little pixel, artifact it takes away from the purpose of the 7d. The 7d is truly amazing camera for 1900 bucks (with standard lens 28-135). It offers multiple framerates, fixed the audio to 16 bit from the crappy 12bit of the 5d. Not to mention the obvious shallow dof, amazing low light, and full manual control. The 7d can beat cameras that are 10 times as expensive in alot of areas, just like little canon hv30's and hfs100's beat prosumer cams in the 2k-5k range (pq wise the hv30/hfs100 beats the crap out of the dvx). I have an xl1s and went to an hfs100 and i am amazed by its results. It can match the sharpness of the xh a1 and other cams easily. Anyway i may want to add the canon 7d for its much much much better low light than my hfs100 (which honestly is mediocre but for the price of 900 dollars an amazing cam), and much much shallower dof. I also love the colors from the 7d (at my work we use the 5d mark II and luv it but couldn't really afford it). Hopefully the 7d price will go down in the next 3-6 months. Remember you don't need a 100,000 dollar camera to do quality work. Ppl have done amazing things with little hv30's and hfs100's, 5d mark II etc. A little compression issue that is honestly not visible at all is not going to hinder your creativity.

The 5D2 is 16bits at 44.1kHz. Right now the 5D2 is superior for audio because of Magic Lantern. Once Magic Lantern is ported to the 7D, the 7D will have a small edge with 48 kHz, but that's a very small edge indeed.

Personally, I compose film music, like most composers who can't afford a live orchestra, with a sampler. 99 percent of sample libraries use 44.1 kHz audio. Because of that I do all my post work at 44.1 kHz at 24 bits. At the end I resample the whole thing to 48 kHz and dither to 16 bits for the final product. For me 44.1kHz is actually superior in this regard, but I won't complain about 48 kHz.

I think the old DAT recorders used to record to 48 KHz.

Let's keep things in perspective... It's a consumer camera with a 1/8" microphone jack and questionable A/D. 44k vs 48k is really meaningless in that context. If you want quality, the answer is of course to record on a real audio deck.

Chris.


We have had a bit of a play with the Galbraith footage over here and attempted to convert the 30P to 25P by retiming the playback slower so that the render (hopefully) remains frame-by-frame. We found that if you do not select frame blend "off" in Premiere Pro, that a stutter developed in the playback and it did not look quite as good. This is offtopic for what you were doing but it might be worth checking anyway.


Ian.


My 1993 Sony DAT recorder still records in 48K and quite nicely but then I am also a bit of a technological luddite. I would probably be still recording to transcription disks with a Neumann cutter if people of my ilk were allowed to have their way.



Thank goodness for innovation and progress. Whatever their limitations may be these cameras remain game changers and they are only first generation of their concept. The future is indeed interesting.

You might be surprised. If you provide the camera a clean, hot signal and turn the gain down, the result is VERY clean, and the sound quality quite good. In my tests, recording with the juicedLink into the 5D with Magic Lantern is noticeably cleaner than I could get with the H4n. At our recent 48 Hour showing, our result was nice and clean, even at theater levels.

I'm not saying it's audiophile quality. The audio is absolutely crisp and detailed, but slightly crunchy. I believe that the filter isn't cutting all of the aliasing. I should buy or design/build a good passive filter and see if that makes a difference.

As long as the 1/8" cable is short and driven hard, the results are surprisingly good.

You can see/hear my reviews here: 1. Canon 5D Mark II Audio Exposed - Boom Mic (juicedLink, Zoom H4n, Microtrack II, BeachTek) on Vimeo

Agreed...amen to dat...I meant.... "that." :)