NEX5 video encoder comparison to GH1

Vitaliy Kiselev

Vitaliy Kiselev

Carbonite Member
Look here:

1273624733.jpg


Almost camcorder format.
CABAC
Reframes : 2 ( 1 for GH1)
B Frames present, but not so widely used as in camcorders (reason - 2 reframes only).
More intellect in I,P,B frame size variation (GH1 is dumb here).

But why they choose so low P and B frame sizes, and spend so much on I frames (in motion this changes, btw, more is spend on P frames)?
From review samples you can see overcompression in motion sequence, so bitrate is still low for such encoder. Something about 30Mbit will be ideal.

Main conclusion - this is first EVIL camera with good encoder build in.
If manual mode could be enabled and bitrate up to 30Mbit it'll produce picture much more clear than Canon guys.
 
If manual mode could be enabled it would be amazing.
I've read everywhere that this camera only shoots interlaced 30fps, but today I've downloaded a couple of clips from from a well known spanish review website and the clips were 25p! I've started looking around and apparently this camera can do 1440 x 1080 @ 29.97 / 25 fps if you're recording MP4. I'm not sure if this a PAL thing only.
 
pal or ntsc

pal or ntsc

Pal is 25p and ntsc is 30p(29.97 or whatever it is;-) ) but it is mjpeg only 1080i is avc.
 
Yes.
This baby with good manual lenses, hacked firmware with bitrate up to 30-35Mbps, longer GOP and ability to preserve shutter from manual mode will be very-very good.
 
Tester13 I heard that Sony will release new f/w

Tester13 I heard that Sony will release new f/w

Because the 3d aspect needs enablng and wont be available at launch, also they have some new 18-200 lense like the GH1 14-140 improved OIS so that may need a f/w update.

In your opinion better to work from NEX or GH1 for better video, could the mjpeg 1080p is subsampled o 1080 to the 1040 for some reason and not very high bit rate.

The NEX is dam ugly with the lens though, but hey of it delivers so what I suppose.

One last thing it may help if you fit the ND filter to the NEX to get the shutter down, I think alot of footage is in daylight with very very high shutter speeds.
As you said elswhere fixed at 1/48, 1/50, 1/60 would be real cool, only need six shutters really them and 2x?
 
Tameside said:
Because the 3d aspect needs enablng and wont be available at launch, also they have some new 18-200 lense like the GH1 14-140 improved OIS so that may need a f/w update.
Click to expand...

May be.
Be almost all latest firmware images are encrypted.

In your opinion better to work from NEX or GH1 for better video, could the mjpeg 1080p is subsampled o 1080 to the 1040 for some reason and not very high bit rate.
Click to expand...

Looks like we have interlaced video with heavy Canon-like aliasing and good AVCHD encoder vs GH1 much better aliasing and rolling shutter plus crappy encoder.


One last thing it may help if you fit the ND filter to the NEX to get the shutter down, I think alot of footage is in daylight with very very high shutter speeds. As you said elswhere fixed at 1/48, 1/50, 1/60 would be real cool, only need six shutters really them and 2x?
Click to expand...

Only two shutter speeds are actually required - 1/50 and 1/60.
 
zoom looks good on the nex5?

zoom looks good on the nex5?

No exposure bumps but this could be because they automatically stopping down a fair bit, however, seems as if there's good control on zooming?
 
tester13 said:
I believe that it have complete binning (horizontal and vertical).
In other words it is same as whole sensor image downscaled by simple (i.e. bad) algorithm.
But I may be wrong.
Click to expand...

I've discussed previously what it would technically take for pixel-binning to occur on a NMOS sensor such as the GH1. Its far more complicated then downscaling the whole sensor image.

While I won't say conclusively, for the GH1 to do pixel-binning would be a very expensive investment.

On the subject of how the AF-100 could lower its resolution from 12MP to 1080P. Obviously the easiest is line-skipping. Another would be to capture the whole frame and then scale it, the third way is pixel-binning.

Pixel-binning operates very differently from a CCD and CMOS sensor.

For CCDs, they operate in a binned mode natively. Charge from adjacent pixels in a column are summed in the horizontal register. Charge from adjacent columns are summed at the output node. The effective resolution of the CCD is decreased (in this case to 1080p) depending on how many pixels are summed in each direction.

Also, the S/N (signal-to-noise) ratio is increased by the square root of the number of pixels binned if the noise is dominated by shot noise and the S/N is increased linearly with the number of pixels binned if the noise is dominated by read noise. In a CCD, unlike a CMOS, the binning operation is noiseless since summation takes place in the charge domain. This is a major benefit for CCDs over CMOS sensors. However, CCDs have difficulty scaling-up, something the CMOS is good at, and where pixel-binning would be most beneficial.

Now, if pixel-binning is to happen without a CMOS sensor, there needs to be a fundamental operational change. The sensor would need to integrate an active-pixel array with a passive memory cell array to permit frame-transfer operation. This is so that charge integration amplifiers at the bottom/horizontal of each column of the sensor can allow for binning of the pixel during the read-out process.

So if this sensor is going to do pixel-binning it will require additional passive memory on the sensor to be able to do it. However, I'm not sure the cost-to-benefit ratio is that significant (it would be worth it for video, but not for stills). Hitherto, these sensors have been designed for stills first, with video as an afterthought. Obviously, as the market for FourThirds sensors has shifted, so does how its designed. So there is a possibility that these sensors can be read-out fairly fast with video in mind. Obviously, Panasonic is trying to find a middle-ground between video/stills for these FourThirds sensors.

Also, Panasonic uses an N-type MOSFET sensor (NMOS), and not much is openly published on it, but it seems to be a fairly flexible resolution sensor judging by past iterations, and the aliasing issues haven't been incredibly bad. I'm sure they've made significant strides for a video-centric device AF100 even if it line-skips. At this point its wide-open, but there are a lot of options that Sony/Panasonic (and Canon hopefully soon) could take.
Click to expand...

http://www.dvxuser.com/V6/showpost.php?p=1960924&postcount=73
 
As for CCD it is all clear.
And CMOS or NMOS it is not so.
Main problem is not sensor lines read speed, but rather otherall read speed.
Panasonic sensor have some embedded thing to process full readout to requested size and transfer result using available channels.
Canon skip lines to archive same thing (reducing overall stream to available channels bandwidth).
 
tester13 said:
As for CCD it is all clear.
And CMOS or NMOS it is not so.
Main problem is not sensor lines read speed, but rather otherall read speed.
Panasonic sensor have some embedded thing to process full readout to requested size and transfer result using available channels.
Canon skip lines to archive same thing (reducing overall stream to available channels bandwidth).
Click to expand...

If you're going to 'downscale' the 'whole sensor image' then it means that it must acquire the 12MP image 24 or 30 times a second and downscale it; it needs to downscale it from something after all. If the GH1 could that then we would be having a different conversation.

Now, if you're going to bin pixels so that each photosite isn't a pixel, but rather a group of photosites are read as a single pixel, then like I said before it would require additional passive memory. I'm not sure if this is the 'some embedded thing' you are referring to.

I'm not trying to provide any answers here, there is very little published about Panasonic's "Live-MOS" sensor- it very well may be doing pixel-binning (which is the possibility I referred to in the AF100 thread). I've said before that aliasing issues of the GH1 are comparatively minimal to other sensors, but that may be because its a smaller sensor with a lower resolution. But if the GH1 was doing pixel-binning there should be virtually no aliasing at all, as well as a noticeable boost in S/N ratio in video compared to stills.
 
Steve Castle said:
I'm not trying to provide any answers here, there is very little published about Panasonic's "Live-MOS" sensor- it very well may be doing pixel-binning (which is the possibility I referred to in the AF100 thread). I've said before that aliasing issues of the GH1 are comparatively minimal to other sensors, but that may be because its a smaller sensor with a lower resolution. But if the GH1 was doing pixel-binning there should be virtually no aliasing at all, as well as a noticeable boost in S/N ratio in video compared to stills.
Click to expand...

Aliasing will be still strongly present if you are using dumb algorithm to downscale image.
In reality Panasonic sensor must have some embedded logic that allows to read line, trim it, and combine remaining RAW values (so, only 2x, 3x, etc reduction).
He can also combine lines after all processing.
And send to LSI only resulted RAW image.
This is my current understanding of Panasonic code and testing reults that we have.
As for S/N ratio, you have different sensor operation mode (more heat, higher noise), no noise reduction.
 
I agree with tester, the aliasing and edge artifacts in the NEX5 footage that dpreview posted to youtube are very severe.
 
You must log in or register to reply here.
Back
Top