PappasArts
Veteran
Barry,
Not to long ago you had the hypothesis that the GH1's detail should be turned up to assist the codec. Has this idea changed since your testing?
.
GH1, meet the DSC MegaTrumpets res chart
- Thread starter Barry_Green
- Start date
Barry,
Not to long ago you had the hypothesis that the GH1's detail should be turned up to assist the codec. Has this idea changed since your testing?
.
Barry_Green
Moderator
Not at all. The GH1's sharpening seems quite mild, and it does seem to help keep the codec from mushing together areas that should be showing fine detail.
Sharpening doesn't affect the aliasing one bit, and I like the GH1 with some sharpening added to it.
PappasArts
Veteran
Very good...
Hi Barry,
Yesterday I tested the GH1 with the stock lens, the 20 f1.7, and several of the latest Leica M lenses. I shot RAW stills and 1080-24 video. I took a quick look at them last night, and it appears the 20 f1.7 is sharper than the stock zoom and the Leica lenses (on the GH1 body) in the converted Tifs. I also looked quickly at the Neoscene conversions, and the 20 f1.7 appeared noticeably sharper than the stock zoom lens. The video appears less sharp than the Tifs, but I did not bring it into FCP and apply sharpening or other corrections.
The other thing I recall is the Tifs from the Leica images, while less sharp, had nicer, better saturated color than the Tif from the 20 f1.7. I used the same color temp and other settings in the Light Room conversion, so I am not sure why I am seeing this. I did shoot the scene with a Nikon D3 and 24-70 f2.8 at 70mm and f2.8. The GH1 20 f1.7 (shot at f2.2) was sharper, but the color from the Nikon image is far better (richer, more saturated). All this is a bit rough, coming from a quick look late last night, and I did not try to optimize each image (they all were converted from RAW to TIF using the same settings).
Why is there so much difference between the quality of the still file and the video file? Is this a hardware or software issue?
What does video from an EX1 or HPX300 look like next to the same scene from the GH1?
Thanks,
Bill
Yesterday I tested the GH1 with the stock lens, the 20 f1.7, and several of the latest Leica M lenses. I shot RAW stills and 1080-24 video. I took a quick look at them last night, and it appears the 20 f1.7 is sharper than the stock zoom and the Leica lenses (on the GH1 body) in the converted Tifs. I also looked quickly at the Neoscene conversions, and the 20 f1.7 appeared noticeably sharper than the stock zoom lens. The video appears less sharp than the Tifs, but I did not bring it into FCP and apply sharpening or other corrections.
The other thing I recall is the Tifs from the Leica images, while less sharp, had nicer, better saturated color than the Tif from the 20 f1.7. I used the same color temp and other settings in the Light Room conversion, so I am not sure why I am seeing this. I did shoot the scene with a Nikon D3 and 24-70 f2.8 at 70mm and f2.8. The GH1 20 f1.7 (shot at f2.2) was sharper, but the color from the Nikon image is far better (richer, more saturated). All this is a bit rough, coming from a quick look late last night, and I did not try to optimize each image (they all were converted from RAW to TIF using the same settings).
Why is there so much difference between the quality of the still file and the video file? Is this a hardware or software issue?
What does video from an EX1 or HPX300 look like next to the same scene from the GH1?
Thanks,
Bill
Thanks!
Thanks!
Ditto on that. Thanks for these comparison tests- very, very helpful.
Thanks!
Ditto on that. Thanks for these comparison tests- very, very helpful.
It's a hardware issue, all the current DSLRs (and GH1) have to either bin their pixels - considerably - or subsample a handful of them (or both) to get their video images. When you do this you take a major resolution hit. The hardware simply isn't fast enough - yet - to be able to scan the entire sensor pixel by pixel for its video image the way it does for a still image.
An ideal video sensor for a sharp 1080p would only need about 10mp but would require a bus/processor speed faster than a 7D.
Barry_Green
Moderator
Because these are still cameras (all of them, the 7D, the 5D Mark II, the GH1, the K-X, the D3xx, whatever). They're all primarily still cameras. Their video performance is a majorly compromised add-on. None of these cameras are actually designed to deliver good-quality video.
Look in the "Canon 7D, meet some peers" thread and you'll see where I shot footage from a 7d, an HMC40, and the 7D in stills mode. In stills mode, resized for video, the 7D is a million times better than the 7D in video mode. And it's the same for all these cameras.
Either of those video cameras will blow a DSLR into the weeds, as far as image clarity and dynamic range. The DSLR will have an advantage in shallow DOF.
Stephen Mick
Totally Usable
And in low-light conditions as well, even though my EX1 is pretty good in low light.
Ben_B
Veteran
I've noticed that despite inferior downscaling to say photoshop, part of what makes the video from these cameras look so good in real world applications is the fact that they have a large sensor that is then downscaling, even if it's doing so poorly. And this is only gonna get better 
PappasArts
Veteran
Barry,
Have you tested all the film mode curves with your charts?
Have you tested all the film mode curves with your charts?
Barry_Green
Moderator
For dynamic range? Or gamma, or ? No, but I was going to run them all through on the Stouffer for practical dynamic range...
PappasArts
Veteran
PappasArts
Veteran
Barry I just saw your 7D Stouffer tests. Excellent!
When I are you going to do the GH1 with the Stouffer?
.
When I are you going to do the GH1 with the Stouffer?
.
Barry_Green
Moderator
Today.
PappasArts
Veteran
Awesome!
Psynema
Well-known member
I have my GH1 back, and am now able to start duplicating some of the tests the 7D got put through.
First up, is the DSC Labs MegaTrumpets 12 res chart. For those who didn't follow the other thread, basically the MegaTrumpets 12 is designed to handle up to 4K cameras like the Red One or Dalsa Origin. My older res chart topped out at 700 lines, not nearly enough to show what these cameras can do, so I got the MegaTrumpets 12. It's great because its highly precise and it shows resolution trumpets not just on the horizontal and vertical, but also at many different angles.
As expected, there are mixed results, and the overall results show the same basic kinds of patterns that the 7D exhibited. Here's the chart, side by side with an extraction of the 7D's chart:
I've gone through and marked some areas of special interest on the chart. At first glance, the GH1's chart looks a lot better than the 7D's, because the GH1 doesn't have the bizarre color mush that the 7D exhibits. However, other than that particular anomoly, they're fairly comparable.
In spot "A", the Canon looks like it's actually rendering cleaner than the Panasonic. There's not much horizontal resolution here, about 700 lines on the Canon, and maybe 600 on the Panasonic before the aliasing renders it moot to look further.
In spot "B", that's where the Canon just pukes in an alarming fashion, turning a pure black & white chart into a mishmash of purple and green. The Panasonic does much better. The Canon doesn't even resolve 500 lines cleanly, it's already heavily aliased even at the start, whereas the Panasonic gets to probably 550 before the aliases blur the res completely. As far as detail extinguish, the Canon gives up at about 850 lines, whereas the GH1 is still rendering aliased detail until about 1100. This is not actually good, it'd be good if there was clean rendering until 1100 lines! The actual and legitimate detail is higher on the GH1 than the 7D vertically, and higher on the 7D than the GH1 horizontally, by probably about the same margin. Based on that, I predict both cameras will deliver images that are about the same actual sharpness. The color patterns are disturbing on the Canon however; I suspect it will cause problems on greenscreening that the GH1 wouldn't have.
In spot "C" there's some interesting things happening. On the Panasonic side there's a tiny bit of purple color contamination, never a good thing. Nothing compared to the Canon's color contamination in "B" or "D", but still, it's there and it shouldn't be. Look at the pattern of lines on the Canon side in "C" -- they're actually sweeping fully horizontally! Once you see the pattern on the Canon, look at the Panasonic side and you can see that it's doing it too. That's bizarre, and that's why aliasing is so bad. It makes you think you're seeing something, but it really isn't there! The charts consist of straight lines from the outside to the inside, straight nearly-parallel lines that converge at the center of the chart. So it looks like the Canon and Panasonic are rendering sideways lines, but it's supposed to be perfectly straight lines! What looks like detail here is "false detail", it's contamination in the picture. A lot of the perceived sharpness in the GH1 and 7D and 5D can be directly attributed to this kind of false detail.
I mean, let's look at the HMC40's trumpet so you can see what a trumpet is supposed to look like:
That's what it's supposed to look like. Clean line separation until the lines blur together into mush. And once they blur together, they should stay together. All the wedges should look like this. The way the 7D and GH1 are performing, there's all sorts of false detail being rendered. That false detail is a just plain wrong representation of the picture!
Okay, moving on, to "D" you can see the Canon again exhibiting some bizarre aliasing and color mushing. Keep in mind that there should be 15 thin black & white line pairs in this image, and only 15 thin black & white line pairs. Where did it come up with four purple & green lines? This is not image detail, this is contamination.
On "E" you can see a reasonably well-behaved trumpet from the GH1. Clean detail, little aliasing, nice and cleanly rendered down to detail extinguishing at about 900 lines. But -- then what happens? We get a return of some bogus false detail afterwards... sigh.
On "F" you can see really bad examples of this bogus false detail. After the image has completely exceeded the sensor's ability to render it (the complete detail extinguish point), then some "detail" returns, and we see more lines. These lines are fake. They are wrong. There should be 15 lines there, not four. And they should converge inwards, not outwards. These fake alias lines are bending away from the converge point, not towards it. The GH1 does it too, just not as bad. This is pollution in the image. It can actually make the image look sharper, but it's cheating, an ideal camera design would filter all that stuff out.
Net conclusions? The GH1 and the 7D (and the 5D) don't display true high-def images. The GH1 and 7D look like they probably have about the same amount of real detail in 1080p mode, which would be the amount of detail you could expect from a 720p camera. They do, however, render a much sharper-looking image because of all the false detail they let through. Even though their overall true detail might be comparable, the GH1 is clearly much better behaved on the charts, showing less aliasing and less bothersome color contamination.
Both cameras are capable of truly astonishing imagery. But both of them are resorting to smoke and mirrors to make it.
As for 720p, it's no contest -- while they both have aliasing, the GH1's chart just looks much better. Substantially higher res in the vertical; the 7D is well into the green/purple mush where the chart even begins, whereas the GH1 looks to be clean to 550 lines or more. The 7D's horizontal resolution is strangely well behaved, compared to all the other trumpets... odd.
Uhm, what lenses were these tested on? I hope they got the same glass with adapters, because how do you separate camera performance from lens performance?