5D MKII samples by Vincent Laforet

Discussion in 'Canon EOS' started by hawkman, Sep 28, 2008.

  1. make of them whatever you will,
    http://www.usa.canon.com/dlc/controller?act=GetArticleAct&articleID=2128
     
  2. Interesting, they all look soft or slightly out of focus. Wonder if that's a problem with the autofocus not working correctly or of Canon using too strong of an AA filter.
     
  3. Are tese actual photo captures? or capturres from the video mode?
     
  4. I checked a couple of shots. The TS shot and the 50/1.2 shot have a lot of camera movement in them making them very soft. A lot of the shots are in low light so I suspect many of them could be soft. The 85/1.2 L shot has extremely narrow depth of field and focus is somewhere on the chin (centre of face) rather than the eye. So eyes and ears are soft but stuble on chin is very sharp. All of these images are poor for evaluating sharpness. They may be useful in evaluating bokeh but I have already lost interest in the subject matter.
     
  5. "Are tese actual photo captures? or capturres from the video mode?"

    These are 21 mpixel still images, video stills are about 2 mpixel.
     
  6. "too strong of an AA filter." I doubt it, as they usually go with a weaker AA, I think it is the SW noise reduction that is washing out some of the detail, comparing psoted 50D RAWs and JPEGs looks like Canon updated their image processing engine with D!GIC4 which now includes rather strong noise reduction compared to what they had before...
     
  7. Not a camera test more a lens test.
     
  8. I`d agree John, lot of camera shake, 1/5th sec wide open another 1/20s, nothing to benchmark here :)
     
  9. I agree entirely. I have followed Vincent's work on Reverie with the 5D2 all the way through. The shutter speeds on these examples are way too low to be meaningful samples for evaluation. The guy was making a film under immense working pressure, taking photos were the last thing on his mind at the time. The movie was utterly exceptional and shoes the powerful capability of the 5D2 but we need to look elsewhere for good sample photographs.
     
  10. "lot of camera shake, 1/5th sec wide open " I think he was using a tripod, otherwise the photo would be blurry as hell.
     
  11. `I think he was using a tripod, otherwise the photo would be blurry as hell.` Hi Arash, I`m not sure what you mean on blurry as hell. I just grabbed an old 20d and an ef135 SF lens, its almost dark, 100iso 1/4sec f2.8 and took a pic of my front door 20+ feet away in std jpeg, hand held and is as expected a lil soft (just normal USM for AA filter.:)
    00R0Oa-74311584.jpg
     
  12. 100% crop but at 300% the wire seperation is quite clear, I am having an issue with the lens after a rebuild, I think it needs recalibrating. and I did not work out DOF but is this what you mean as blurry as hell? Thanks
    00R0Oi-74313584.jpg
     
  13. Don`t get me wrong Arash, was only slow shutter shots I mentioned could be shallow DOF with others, only so much Vincent could in the short time he had it

    Cheers
     
  14. At large apertures such as these, image detail is typically limited by the lens, not the sensor. They look good to me.
     
  15. These shots illustrate why technique is as important as the gear you use. Any modern DSLR camera mounted on a tripod could have produced similar or superior results with good technique.
     
  16. Is it possible that this camera can capture 21 MP stills during film shooting? A cool feature but maybe that's asking too much. In any case it would explain some of the blur since all images would then be manually focused.
     
  17. Chirs JB, your photo is very sharp you must have steady hands, yes it is possible to handhold the camera for 1/4 sec.
     
  18. "why technique is as important as the gear you use"
    I am not sure if you are in a position to question Vince Laforet's technique, do you know who he is?
     
  19. The near absence of fine-grained noise patterns @1600 in these images point to a rather effective noise reduction. Packing >20MP on a full-frame sensor naturally produces more noise compared to for example 12MPs. At the guy's hair (img #6) you can see blob-like color artifacts...
     
  20. Just discovered a "soft" red hot pixel over his ear (again image #6)...notice that it is a rather *soft* hot
    pixel (camera shake and lens issues naturally cannot account for that of course) - conclusion?!?!
     
  21. The samples at dpreview.com are much better. Even those, sharp as they are, are improved by minor applications of unsharp mask. Resolution and sharpness are not the same thing.

    --Lannie
     
  22. Anders: Yes, you can shoot at full resolution whilst recording video on the 5D2.

    By the way, look at the shot of the woman's face in silhouette... that one isn't blurred.. he got lucky with that one.
     
  23. Sorry again - in the image with the red girl this soft red pixel (and another white one) is also present, look also at her dress, shouldn't it be smooth...? I just did a quick spectral analysis across a region of the image which seemed well in focus (see attachement). Attached is the whitenend amplitude spectrum of the green color channel (middle), the other channels look similar except being flatter (original patch 512 x 512 pixeles, DC-part in the image center, whitening by variance, application of a 4-term Blackman-Harris Fourier window to reduce tiling artifacts). The spectrum on the right was computed from a typical normal photograph shown for comparision. As you can see, in the (middle) Canon spectrum there are few other-than-blue colored regions (=OR) beyond the center part indicating the absence of in-focus-details. In the right spectrum shown for comparison, there are ORs at all frequencies (albeit orientated). The Canon spectrum is relatively flat, there is indeed few detail in these images. For me the spectra smell like noise-reduction (speculation!), not so much like motion blur.
    00R0iq-74455684.jpg
     
  24. Any modern DSLR camera mounted on a tripod could have produced similar or superior results with good technique.
    This isn't quite true. If you look at the exposure data, it becomes obvious that at those apertures and shutter speeds, with this kind of lighting, considering the potential for subject movement (ie. slower speeds could be used), these are very good results. Obviously they do not demonstrate the sharpness potential of the camera, but they do illustrate that it can produce remarkable results in adverse lighting.
    The samples at dpreview.com are much better
    Again, not true. The dpreview.com samples show sharp images, yes, but they're worthless junk as photographs. Vincent's images are excellent, though not illustrating sharpness.
     
  25. "Again, not true. The dpreview.com samples show sharp images, yes, but they're worthless junk as photographs."

    I was addressing the issue of softness raised above, especially near the beginning of the thread.

    --Lannie
     
  26. These images are captured form the Movie he shot. He stated that when he originally started. These are just screen
    captures from Reverie. He stated he had no still shots to show. Also, these are also described as in camera JPEGS
    so they would naturally be softer than usual. Everyone, dont jump the gun just yet. I'm sure once someone gets the
    camera and does some real "Still" image tests, it will come back with very good results. I've seen some from the
    1DsIII that are unbelievably sharp. But they were shot with the best lenses and good enough shutter speeds and
    f/7.1 aperture to tell. I'm sure the new sensor will be that good.

    Look at this image. I dont know how it will post, but you can copy it and it will work.

    http://ftp.robgalbraith.com/public_files/EOS-1Ds_Mark_III_Coach.jpg

    The detail in the eyes and mouth area are absolutley breath taking. SHARP! But the lens used made the difference.
    The lenses Vincent used are good, but not as good as a 500 f/4. The lenses he used are close to wide open and are
    all noted for being a little soft in that range. The 500mm is sharp all over. And you can see what that sensor is
    capable of. Everybody, just wait. Bottom line, we know it will produce the best results you can buy in 35mm format.
    If you need more, you need about $40,000.
     
  27. "These images are captured form the Movie he shot"

    Wrong, these are still images, movie is 1920X1080 pixels, a still frame from a movie is highly compressed and useless. You should also note that when you capture a still, i.e. press the shutter bottom during movie capture, camera stops and takes a normal photograph. no difference in quality whatsoever.
     
  28. Just quoting what he said in his initial test. That all images shown on his site are movie captures. I understand what the movie resolution is.
     
  29. David,

    These are indeed actual photographs, not movie captures. How else did he quote the shutter speeds of 1/15 sec etc? I remember reading on his blog that he was taking a few photos during the shoot.
     
  30. "Just quoting what he said in his initial test. That all images shown on his site are movie captures. I understand what the movie resolution is."

    I think you are confusing two separate case

    1) A still capture from a movie, i.e. basically pausing the movie and saving a frame, the will be a heavily compressed 1920X1080 image which will also suffer from motion blur artifacts etc. this is obviously not the case here and it will be of no use

    2) pressing the shutter button to capture a still when camera is recording a movie clip, in this case camera will take a normal photograph as intended just like normal operation mode.

    Whether Vince captured stills during shooting a movie clip or not it makes no difference in terms of IQ
     
  31. Arash> I know how all of this works. What I'm saying, he said that he was posting images from video capture. I also
    see some of those images posted here. OK, maybe not all of them, but I do see some. No need in explaining
    something so easy to understand.

    And yes it does matter. You can grab stills out of a movie clip(already taken), which will indeed have lower resolution.

    This wasnt my biggest point anyway. My point was more in relation to lens choice to effect sharpness. Regardless
    whether he's shooting a movie or still's, the lens choice wasnt the best for showing what the sensor is capable of.
    The softness you see is "In-Camera" JPEG and lenses in the soft aperture range. Not sensor or even NR issues. I
    have seen much better results at same ISO's using 135 primes etc. These 50 1.2L/85 1.2L are known for softness
    under f2.8. There's no arguing that statement. I'm not saying they are not good, but can not shown sensor capability
    in that range..
     
  32. "And yes it does matter. You can grab stills out of a movie clip(already taken), which will indeed have lower resolution."

    David, I am not going to debate this issue any further, if you really think these 21 mpixel images are stills from a video clip I have nothing more to add-I am sorry.
     
  33. Keil,
    Good job with the FFT spectrum, the frequency response indicates spectrum has been equalized, due to processing or NR. you can run FFT on red, green and blue channels seperately and find there is strong corrolation factor, indicating chroma NR. but this is obvious by just looking at the pictures, there is no noise in the dark uniform areas, these areas will normally have noise due to nonuniformity in sensor dark current, so there is NR for certain and it has taken its toll on low contrast detail, but the counter argument is that users of this camera will most likely shoot in RAW so we have to see how much detail DPP can recover from RAW, I don't care about JPEG that much.
    BTW, did you write the matlab code yourself or is this a standard toolbox?

    Regards,
    Arash
     
  34. Hi Arash,

    thanks...yes, the whitening algorithms were devised by me and previously used to connect natural image statistics
    with face image processing in humans (described in Proc Biol Sci. 2008 Sep 22;275(1647):2095-100, or
    http://arxiv.org/abs/0804.0759).

    To all:

    It was suggested above that the softness derives from lens softness...however, I also mentioned above that I
    discovered a hot pixel in the sample images which was blurred, and this points to noise reduction. I usually do
    not use NR, and the hot pixels I saw in my cases were crisper. Furthermore, in the black areas of the images one
    should at least see some fine-grained noise at iso 1600 if there were no NR, but instead one sees film-grain-like
    structures and big (blue) color blobs. This also could be interpreted as being a consequence of NR. The spectra
    show also that there is nearly no high-frequency content in the images...so where has the noise gone? Of course
    these are speculations, but at least I have threefold evidence to support my claims.
     
  35. It's a 5D with some modifications. Why would you need to look at sample photos? You know exactly what they are going to look like...any other photo.
     
  36. Hello again,

    >Look at this image. I dont know how it will post, but you can copy it and it will work.
    >http://ftp.robgalbraith.com/public_files/EOS-1Ds_Mark_III_Coach.jpg

    Yes, this indeed seems a good & sharp image. It was taken at ISO200 (according to the image legend), and in this case no NR is necessary. Nevertheless, for pixelists like me, in the hair you see nice interpolation artifacts (from interpolation of the missing colors from the Bayer pattern), but I think at the moment this simply represents the state-of-the-art in image processing...at least I never saw benchmark images with hair with the latest published algorithms :)
     
  37. MS, this is what I was talking about. Its all in the lens. Its as good as it gets. Complaining about the IQ from that shot(the one linked) is pointless. You wont find better in 35mm format. For those that find this is not good enough, there's no way to make you happy.
     
  38. Keil: "I just did a quick spectral analysis... application of a 4-term Blackman-Harris Fourier window to reduce
    tiling artifacts)."
    <p>Arash: "Keil, Good job with the FFT spectrum...noise due to nonuniformity in sensor dark current, so there is
    NR for certain and it has taken its toll on low contrast detail...BTW, did you write the matlab code yourself or
    is this a standard toolbox?"
    <p>Keil: "...the whitening algorithms were devised by me and previously used to connect natural image statistics
    with face image processing in humans"
    <p>Eureka! A spectral analysis coded by a face recognition software expert was just the final piece I needed to
    complete the
    answer to my quagmire (5DMkI vs 5DMkII). I've now made up my mind to go for a 5DII based on your results ;-)
    Sheesh! Here I was thinkin' this is just a humble photographers' forum :)
    <p>Mark :]
     
  39. "Here I was thinkin' this is just a humble photographers' forum :)"

    It is a photography forum, however image processing is an integral part of digital imagery, I'm sure back in the film days people had discussions about chemistry of films and development processes, today, it is electrons and image processing that constitute the basis of digital photography.
     
  40. `Eureka! A spectral analysis coded by a face recognition software expert was just the final piece I needed to complete the answer to my quagmire (5DMkI vs 5DMkII). I've now made up my mind to go for a 5DII based on your results ;-) `

    I`m sorry but I`m at a loss here, how and why would this type of info influence whether you would buy one body over another and at what point would you expect to see a difference? I owned a pro lab and I`ve never heard this sort of talk, tho grew up with film the main analysis was chem & temp, and with photography was the affects of varying light and its effecting the layers of emultion in different films. but most folks just took pics :)
     

Share This Page