Barry_Green
Moderator
How is the GH1 a "720p native" cam? It has a 10-megapixel sensor. I'm not following what you're saying here at all.
My thoughts on sensor size
- Thread starter Daniel L.
- Start date
My bad.. I meant to say GH1 a camera that can shoot 720P natively. It's utilizing the full width of the 10MP sensor and downsizing the image to 720P (plus other data reducing algorithm it employs)
In other words, given the subject of the same size. It is always better to have that subject fill as much of the entire frame as possible(to utilize as much of the sensor as possible) versus using a higher resolution sensor and have the subject fill a smaller area of the frame with the intention of cropping the resulting image to achieve the same FOV.
Doesn't that only matter if you plan on going back to 1080?
If it's a straight rip out of the frame and you don't plan to uprez then it should be fine. Naturally, in a versus scenario, a 1080 filled fre would look more resolute than a 720 matched to that 1080.
I also think an mkii crop to match the gh1 720 would still be more resolute. That's a guess though. Dunno for sure.
I'm on location right now shooting but I'm going to start experimenting with this before I sell.
If it's a straight rip out of the frame and you don't plan to uprez then it should be fine. Naturally, in a versus scenario, a 1080 filled fre would look more resolute than a 720 matched to that 1080.
I also think an mkii crop to match the gh1 720 would still be more resolute. That's a guess though. Dunno for sure.
I'm on location right now shooting but I'm going to start experimenting with this before I sell.
Barry_Green
Moderator
But... do we know that? How do we know that it's not reading the entire frame, and scaling down? Has this been empirically proven somewhere yet?
This does seem like a logical assumption, and typically on CMOS cameras you do get faster frame rates by windowing the sensor (hence why Red's 2K can run at faster frame rates than its 4K). But... again -- do we know it's doing this? Or is this conjecture? Has some sort of test established that this is the way it's working?
Barry_Green
Moderator
Well, 720p = about 1 megapixel. The GH1 shoots 10mp, and downconverts to either 1080p or 720p, as appropriate, from (as near as I can guess) the full sensor. (as to whether that downconvert is happening due to row skipping, or actual resizing, is as far as I know undetermined yet).
There's no cropping going on, is there? It's been a while since I had the GH1 in my hands, but I don't remember any field-of-view issues happening when changing the various modes; field-of-view changes would be an indicator of cropped sensor usage. In fact, I seem to recall the field of view gets ever so slightly wider when going into the HD modes, which would imply that it's using more of the sensor than even in full-frame photo mode.
Still not sure I follow exactly what you're getting at though. The Canon's sensor is bigger physically, so even though it has twice the pixels, it's also (ballpark-ish as I don't feel like researching up the specs) roughly twice as big, so pixel size should be the same. And therefore, extracting a 1280x720 image out of a 17mm wide crop of a 21-mpix 35mm-wide sensor using 5-micron pixels should be about the same as extracting a 1280x720 image out of a 17mm wide sensor with 10m pixels, assuming 5-micron pixels... should be same-for-same, no? I don't know if the pixels are 5 microns wide, but if one sensor has 2x as many pixels and 2x the surface area, it stands to reason that the pixel sizes would be the same between the sensors, so therefore (using same-size pixels) a 1280x720 extraction from one should be roughly comparable to a 1280x720 extraction from the other, assuming both shots are framed such that the subject is the same size within those 1280x720 frames...
There's no cropping going on, is there? It's been a while since I had the GH1 in my hands, but I don't remember any field-of-view issues happening when changing the various modes; field-of-view changes would be an indicator of cropped sensor usage. In fact, I seem to recall the field of view gets ever so slightly wider when going into the HD modes, which would imply that it's using more of the sensor than even in full-frame photo mode.
Well, okay, I think I see what you're saying here; on a Red One you can put it in "2K" mode and it uses the central portion of the sensor, but obviously the "2k" mode doesn't hold a candle to its full 4K mode. But: if the higher-resolution sesnor is physically bigger, I think it cancels out what you're saying.
Still not sure I follow exactly what you're getting at though. The Canon's sensor is bigger physically, so even though it has twice the pixels, it's also (ballpark-ish as I don't feel like researching up the specs) roughly twice as big, so pixel size should be the same. And therefore, extracting a 1280x720 image out of a 17mm wide crop of a 21-mpix 35mm-wide sensor using 5-micron pixels should be about the same as extracting a 1280x720 image out of a 17mm wide sensor with 10m pixels, assuming 5-micron pixels... should be same-for-same, no? I don't know if the pixels are 5 microns wide, but if one sensor has 2x as many pixels and 2x the surface area, it stands to reason that the pixel sizes would be the same between the sensors, so therefore (using same-size pixels) a 1280x720 extraction from one should be roughly comparable to a 1280x720 extraction from the other, assuming both shots are framed such that the subject is the same size within those 1280x720 frames...
I did say it was my speculation :smile:
In fact, I think it's wrong. Programming cameras in the past, the readout has always been a function of the sensor hardware. It's easy to define a window, but to program a sensor to only read specific fields was never available.
I agree with you on your last point.
In-camera cropping is not possible.
"Hardware adjustments"? Outside of the Canon laboratory, not possible.
You use shorter lenses to adjust for Academy ratio, which means jello is reduced, then you crop, in post. In so doing you've decreased your resolution, increased your depth of field, and retained the same field of view you would have had using longer lenses and not cropping.
Oh right. Not longer but shorter and moving camera back while framing with guides. My bad. Trying to think, operate this big ass red and iPhone type at the same time.
Hmmm. I guess it's a try and see thing. At least for me. Seems like you're still recording the same amount if jello regardless. It just looks like it's less maybe?
Hmmm. I guess it's a try and see thing. At least for me. Seems like you're still recording the same amount if jello regardless. It just looks like it's less maybe?
Barry_Green
Moderator
And that's the question... if they used circuitry to specifically link pixels together, perhaps they could read out a row-skipping/column-skipping version, but... what about the Bayer pattern?
... that's why I was asking if this was known yet. I know people are saying that the 5D skips rows, and it seems reasonable up until the point where you hit the Bayer pattern and then you have to start wondering whether it's a practical technique or not. According to the specs, the GH1's sensor is a >4K sensor (4352 x 2448). 4352 does not evenly divide by 1920, nor does 2448 evenly divide by 1080. How would it just skip pixels to get that?
Now, perhaps the GH1 is using a slightly smaller crop of the sensor, 3840x2160, for motion video. If so, the field of view should shrink for movies as versus 16:9 still frames. But even so, using 3840 x 2160, you'd be tempted to think they could just skip every other pixel, extracting only the even pixels (upper left out of each quadrant of four) to make a 1920x1080 frame. But if using a Bayer pattern over the sensor, that just won't work! You'd end up with a monochrome image. You'd always only ever get green pixels! Or blue pixels. Or red pixels. In the traditional Bayer pattern, the upper-left pixel of any four-pixel quadrant is always going to be the same color, so simple row-skipping isn't going to work.
Unless it's some sort of modified Bayer pattern, where the grid is rotated 90 degrees for each successive 4-pixel block... which would seem to be the obvious choice... hmmm...
RG RG RG RG
GB GB GB GB
becomes
RG GR BG GB
GB BG GR RG
and when sampled at every-other-pixel, you get
RGBG
GBGR
Hmmm... that seems like it'd work, but it'd play hell with the full-res demosaic. Ostensibly you could sample at full res and use a modified Bayer pattern, but what happens to the blocks where there's two greens next to each other, how does the demosaic work with that? Now, this type of arrangement would optimize the possibility of reading out at every-other-column, every-other-row and get alternating R/G/B/G pixels... which would work for accurate demosaic'ing, but also make the aliasing situation appear due to spacing between the pixel samples... but then what about 720p? You'd have to sample every third pixel out of 3840x2160 to end up with 1280x720, but the Bayer pattern would again be totally wrong. If you used the above example, you'd get
RG GR BG GB RG GR BG GB = RRGGBB
GB BG GR RG GB BG GR RG = GGRBGG
Totally wrong.
So does it sample 1920x1080 in row/column skipping format, and then downscale that to 720P? If so, that leaves no provision for the faster frame rate, and in fact seems like it'd take up more processor time and therefore make the faster frame rate less likely, yet we know that it does indeed have a faster frame rate in 720p...
So this is why I ask. I mean, single-sensor cameras with a Bayer or other similar pattern are complex devices to scale... Red One makes a lot of sense because they keep the RGB pattern intact and they window down on the sensor. If the GH1 does that, it all makes perfect sense too. But I don't think it does that, I think it keeps the same field of view, so therefore people speculate about row-skipping, but what kind of row-skipping pattern makes sense for both 720p and 1080p?
Is it some sort of modified RGB Stripe sensor? Or some other pattern?
The specifications only say "primary color filter". Not much help.
Last edited:
If you're doing this comparison in the two camera's still mode then it makes sense..
because the still mode would utilize the full resolution of their sensors.
However a 1080P video from the 5D2 doesn't resolve twice as much resolution than the 1080P from the GH1.. So cropping a small area from the 1080P frame expecting it to be equally as good as the 720P clip from the GH1 is unrealistic IMO.
ydgmdlu
Veteran
No, you're mistaken. Micro Four Thirds is a "quarter frame" format, so in fact the 5D's sensor is four times larger. Thus, the pixels are roughly twice the size of the GH1's.
The video mode on these high resolution still cameras is a great equalizer. 1080P on the 5D2 will show similar resolution to 1080P on the GH1. No doubt about that. So one can't really take the 1080P video from the 5D2, expecting a 720P crop of it be similarily equal in quality to a 720P video shot natively from another DSLR..
