5D4 Sensor Defect Discovered

[Alex_M]

I was told that one can push exposure by +3 stops in DPP at Maximum?? Is that right?

can you do me a big favour and take Dual Pixel RAW (60Mb) shot and process in DPP with DPRAW enabled and see if the fenomenon is still present. I somehow suspect that what we are seing is somewhat related to DPRAW tech. I thank you in advance.

Would using the latest version of camera raw make any difference 9.6.1 I think?

Edit:. Anyone able to replicate with DxO optics pro? I didn't upgrade to the latest version so can't test

I have tested with latest DPP (4.5.10.1) and I also converted the raw to dng. Even 5D3 gives a more tolerable result

 

[tron]

Kevin, please let us know what comes of showing photos to a Canon technician, and what he has to say about all of this

I will! Canon doesn't like saying anything like "defect" or "problem", but they were quick to say what I'm seeing shouldn't happen...much to the dismay of the smartasses that have posted in here. Hahah

When I uncovered a firmware bug in the 1D X and called Canon, they said that shouldn't happen and since they couldn't reproduce it, I should send it in. I declined, worked out the root issue with help from another member of these forums, created a video demo and sent it to Canon, then they could reproduce it.

Point is, I hope they find it's defective, but they may tell you it's within spec.

In this case, you're drawing conclusions based on them having only your verbal description, not even pics. Put another way, if you rush a miracle man, you get rotten miracles.

Neuro or anyone with a 1Dx or 1DxII can you help by trying a similar setup? I know I have exaggerated in the photo and underexposing much is not what I should do but since my 5D3 behaved better (in most part of the photo at least) I want to decide between 1DxII and 5D4 in a moonrise photo behind a temple where similar conditions apply (so as to both preserve the moon surface and the temple). Up to now I revert to taking some photos using Magic Lantern's Dual Iso feature but a better camera would be preferable. In page 10 of this thread I attached two photos of a black jacket, one shot with 5D4 and one shot with a 5D3. I underexposed (manually) 3.5 to 4EV (using evaluative metering and including most of the jacket in the photo).

 

[tron]

I was told that one can push exposure by +3 stops in DPP at Maximum?? Is that right?

can you do me a big favour and take Dual Pixel RAW (60Mb) shot and process in DPP with DPRAW enabled and see if the fenomenon is still present. I somehow suspect that what we are seing is somewhat related to DPRAW tech. I thank you in advance.

Would using the latest version of camera raw make any difference 9.6.1 I think?

Edit:. Anyone able to replicate with DxO optics pro? I didn't upgrade to the latest version so can't test

I have tested with latest DPP (4.5.10.1) and I also converted the raw to dng. Even 5D3 gives a more tolerable result

Right now I have packed the camera and at the same time leaving in a few hours for a small vacation for 4 days. I hope that dealer answers quickly so as to have time to send the camera back.
I cannot use latest raw for now (no photoshop cc or lightroom). But both DPP and dng converter produce the same results... As for +3 I do not know the units used I just test the result by mooving the slider.

Something I noticed. The key issue is NOT how much you move the slider after underexposure but how much you underexpose. In the first case the issue exists in the second it does not.
If I underexpose less photo seems OK.

 

[Sporgon]

But I have to admit these tests are a little extreme. The jacket is black! Even so 5D4 seems to have worse banding (in absence of a more precise term) than 5D3...

What are we supposed to be seeing here?

+1

Looks very good to me on a 4.75 stop lift.

The 5D3 version pushed even more does not have the color bands! Tue in the very dark small parts of the photos it has worse noise - that's 5D3 after all. But you can see the difference in the major part of the 5D3 photo.

That's really weird because I could not see that banding earlier this morning, now I can ! And I wasn't smoking anything last night before you ask

 

[Loswr]

Neuro or anyone with a 1Dx or 1DxII can you help by trying a similar setup?

I could, but given that this phenomenon seems camera-specific, I'm not sure it would mean anything. Especially in my case, since you'd be using a 1D X II.

I know I have exaggerated in the photo and underexposing much is not what I should do but since my 5D3 behaved better (in most part of the photo at least) I want to decide between 1DxII and 5D4 in a moonrise photo behind a temple where similar conditions apply (so as to both preserve the moon surface and the temple).

Temples generally don't move. The moon moves, but not fast enough to matter at an AoV sufficient for a temple and an exposure time appropriate for the moon's brightness). Is there some other reason you are limited to a single shot of the scene? If not, you'd be far better off taking two images and blending the exposures (regardless of the camera you're using).

 

[tron]

But I have to admit these tests are a little extreme. The jacket is black! Even so 5D4 seems to have worse banding (in absence of a more precise term) than 5D3...

What are we supposed to be seeing here?

+1

Looks very good to me on a 4.75 stop lift.

The 5D3 version pushed even more does not have the color bands! Tue in the very dark small parts of the photos it has worse noise - that's 5D3 after all. But you can see the difference in the major part of the 5D3 photo.

That's really weird because I could not see that banding earlier this morning, now I can ! And I wasn't smoking anything last night before you ask

No worries. I have followed you in this forum. I know you are a serious person. As you saw I was not offended. The good thing is that the dealer agreed to pick up the camera (it will happen on Tuesday) and to make some tests with the replacement camera so as to send me a good copy (I hope there is one!)

 

[tron]

Neuro or anyone with a 1Dx or 1DxII can you help by trying a similar setup?

I could, but given that this phenomenon seems camera-specific, I'm not sure it would mean anything. Especially in my case, since you'd be using a 1D X II.

I know I have exaggerated in the photo and underexposing much is not what I should do but since my 5D3 behaved better (in most part of the photo at least) I want to decide between 1DxII and 5D4 in a moonrise photo behind a temple where similar conditions apply (so as to both preserve the moon surface and the temple).

Temples generally don't move. The moon moves, but not fast enough to matter at an AoV sufficient for a temple and an exposure time appropriate for the moon's brightness). Is there some other reason you are limited to a single shot of the scene? If not, you'd be far better off taking two images and blending the exposures (regardless of the camera you're using).

A day (night) after full moon the temple is lighted (usually). Since the moon is not very high, If I use ML I am at the limits of DR. I like shooting one instance although I do understand that there are cases where no matter how good a camera we need to take many photos. Fortunately the dealer agreed to take back the camera. We will see how this proceeds. If the new copy is OK (I am talking about the banding, some noise from Exp Comp is acceptable of course) I will settle with this. Although familiar with just buttons (eos 620,600,rt,1n/1nHS) I prefer the small size and the silent shutter of 5D series.

 

[Sporgon]

But I have to admit these tests are a little extreme. The jacket is black! Even so 5D4 seems to have worse banding (in absence of a more precise term) than 5D3...

What are we supposed to be seeing here?

+1

Looks very good to me on a 4.75 stop lift.

The 5D3 version pushed even more does not have the color bands! Tue in the very dark small parts of the photos it has worse noise - that's 5D3 after all. But you can see the difference in the major part of the 5D3 photo.

That's really weird because I could not see that banding earlier this morning, now I can ! And I wasn't smoking anything last night before you ask

No worries. I have followed you in this forum. I know you are a serious person. As you saw I was not offended. The good thing is that the dealer agreed to pick up the camera (it will happen on Tuesday) and to make some tests with the replacement camera so as to send me a good copy (I hope there is one!)

Keep us updated. I downloaded the file of the graduation (?) from Dropbox that another member had posted, and my version didn't have as obvious banding as his conversion that he posted here, but then looking at his screen grab of ACR I had used a slightly different method to achieve more or less the same picture by moving mid tones more and shadows less.

 

[Labdoc]

Also Labdoc provided a file which doesn't appear to have the issue after a 5 stop push and a 100% shadow push . . . I'd be interested if Labdoc could provide the original RAW file so that I could have a play too.

I went back and tested every picture taken with the MKIV and can not make the banding appear which is a good thing. One pic I expected to have problems, underexposed at 32000 ISO shows only noise but no banding when pushed 5 stops. The fact it happens on some cameras is still disconcerting. I'm not sure if it matters, but I was one of the people who waited on Amazon to deliver, they were at least a week behind all the other vendors, maybe it was from a later batch of cameras. No problem providing RAW files but I have no place to post them and my internet is very slow but still willing to try.

 

[Loswr]

Neuro or anyone with a 1Dx or 1DxII can you help by trying a similar setup?

I could, but given that this phenomenon seems camera-specific, I'm not sure it would mean anything. Especially in my case, since you'd be using a 1D X II.

I know I have exaggerated in the photo and underexposing much is not what I should do but since my 5D3 behaved better (in most part of the photo at least) I want to decide between 1DxII and 5D4 in a moonrise photo behind a temple where similar conditions apply (so as to both preserve the moon surface and the temple).

Temples generally don't move. The moon moves, but not fast enough to matter at an AoV sufficient for a temple and an exposure time appropriate for the moon's brightness). Is there some other reason you are limited to a single shot of the scene? If not, you'd be far better off taking two images and blending the exposures (regardless of the camera you're using).

A day (night) after full moon the temple is lighted (usually). Since the moon is not very high, If I use ML I am at the limits of DR. I like shooting one instance although I do understand that there are cases where no matter how good a camera we need to take many photos. Fortunately the dealer agreed to take back the camera. We will see how this proceeds. If the new copy is OK (I am talking about the banding, some noise from Exp Comp is acceptable of course) I will settle with this. Although familiar with just buttons (eos 620,600,rt,1n/1nHS) I prefer the small size and the silent shutter of 5D series.

You mentioned a "similar situation" to the 3.5-4 stop push. That's going to hurt your tonality, even in the absence of any banding or other artifact. ML dual ISO will cost you resolution.

Full moon landscapes work at golden hour with a single exposure, but at blue hour or later you really should consider blending multiple exposures IMO.

 

[Sporgon]

Also Labdoc provided a file which doesn't appear to have the issue after a 5 stop push and a 100% shadow push . . . I'd be interested if Labdoc could provide the original RAW file so that I could have a play too.

I went back and tested every picture taken with the MKIV and can not make the banding appear which is a good thing. One pic I expected to have problems, underexposed at 32000 ISO shows only noise but no banding when pushed 5 stops. The fact it happens on some cameras is still disconcerting. I'm not sure if it matters, but I was one of the people who waited on Amazon to deliver, they were at least a week behind all the other vendors, maybe it was from a later batch of cameras. No problem providing RAW files but I have no place to post them and my internet is very slow but still willing to try.

Is this not predominantly a low ISO problem, if indeed it really turns out to be a batch problem.

 

[Sporgon]

hi,
a test that you have to do to see the defect is a wide dynamic scene that includes deep shadow as well as bright areas. A real life situation. The quickest way to find such a scene is to find a dark room in your home with a window with some detail and bright sunshine outside. Expose for the scene outside and let the shadows in the room fall where they may . Then try some shadow lifting. If you don't get banding here, then it is not likely you will routinely find this kind of under-exposure in most things you will shoot.


iso 100
this:

Your only problem there is where the camera has recorded nothing in the first place, so you have no problem. Or maybe that's what your saying.

 

[tron]

Neuro or anyone with a 1Dx or 1DxII can you help by trying a similar setup?

I could, but given that this phenomenon seems camera-specific, I'm not sure it would mean anything. Especially in my case, since you'd be using a 1D X II.

I know I have exaggerated in the photo and underexposing much is not what I should do but since my 5D3 behaved better (in most part of the photo at least) I want to decide between 1DxII and 5D4 in a moonrise photo behind a temple where similar conditions apply (so as to both preserve the moon surface and the temple).

Temples generally don't move. The moon moves, but not fast enough to matter at an AoV sufficient for a temple and an exposure time appropriate for the moon's brightness). Is there some other reason you are limited to a single shot of the scene? If not, you'd be far better off taking two images and blending the exposures (regardless of the camera you're using).

A day (night) after full moon the temple is lighted (usually). Since the moon is not very high, If I use ML I am at the limits of DR. I like shooting one instance although I do understand that there are cases where no matter how good a camera we need to take many photos. Fortunately the dealer agreed to take back the camera. We will see how this proceeds. If the new copy is OK (I am talking about the banding, some noise from Exp Comp is acceptable of course) I will settle with this. Although familiar with just buttons (eos 620,600,rt,1n/1nHS) I prefer the small size and the silent shutter of 5D series.

You mentioned a "similar situation" to the 3.5-4 stop push. That's going to hurt your tonality, even in the absence of any banding or other artifact. ML dual ISO will cost you resolution.

Full moon landscapes work at golden hour with a single exposure, but at blue hour or later you really should consider blending multiple exposures IMO.

You are right of course. I admit I do not feel very comfortable with blending although I have used it once successfully (5 stop difference: I was shooting.... you guessed it: moonrise again). But I enjoy much more to just manipulate in ACR raw files.
In the specific case I had used ML in some and not in others. The best result (with difference) was the ML version.

 

[Loswr]

You are right of course. I admit I do not feel very comfortable with blending although I have used it once successfully (5 stop difference: I was shooting.... you guessed it: moonrise again). But I enjoy much more to just manipulate in ACR raw files.
In the specific case I had used ML in some and not in others. The best result (with difference) was the ML version.

As far as exposure blending goes, the moon is pretty simple if it's an isolated element – just cut it out, drop it into the other image (over the blown out moon), and blur the edges a bit. It's more complex if you're shooting from a distance with a telephoto lens and the foreground is superimposed on the moon. But there are ample tutorials on exposure blending, luminosity masks, etc. I'm assuming you have PS, if not it might be more difficult. Regardless, good luck!

 

[ivanku]

Hey Kevin (OP),

Have you gotten any more response from CPS?
I reached out to them and they were fairly dismissive, saying that every camera they produce shows fixed-pattern noise when the files are pushed. I'm disappointed, but I'm keeping my camera and not juggling bodies back and forth to find one that shows less banding. I am curious what kind of response you got from Canon

 

[[email protected]]

So I tested my 5d4 for this banding using a 5 stop push with 100 shadows raising (something I don't come close to needing myself, but this is just to see if the sensor is even capable of that). I do understand that this technique is useful to some, but I do suspect this is pretty rare.

I mostly don't see banding, but perhaps 1 out of 3 shots, I do see some limited banding. I am including the worst of the banding I was able to reproduce in an image below, along with a 1:1 crop.

Thinking that perhaps this sudden need for 5+ stop pushes is not a reasonable standard by which to call a sensor deficient all of a sudden, I wondered how my 1DX would do, it being the much-vaunted flagship camera for Canon just a few months ago. What I discovered was that the noise and banding were consistently much worse on the 1DX, which has the advantage of having a much larger area per pixel. I also include a shot below, taken by the 1DX at a 1:1 crop.

I then looked to see what had the most beneficial effect to fix any banding on the 5D4 in Lightroom. I found that upping the "color" slider under noise reduction helped a little, but the "smoothness" slider took away about 2/3rds of the banding.

Pretty freaking impressed how much dynamic range this 5D4 is giving me, and how many stops of push I can do. I would judge it (by my tastes and standards) to give me 3.5 to 4.5 stops of push. My old 5D3 gave me 2 to 3 stops before I didn't like the results. The 1DX gives me roughly 2.5 to 3.5. This is all pretty subjective in terms of what I consider acceptable, but the relative performance between the three cameras is fairly demonstrable.

The upshot: The 5D4's sensor improved noise and banding enormously, although perhaps a stop or so away from the Sony sensors. I think this thread has been fantastic in establishing how much things can be pushed, but this limit is no more a "defect" than the Nikon D5's fps being a couple less than the 1DX Mark II. The Nikon's frame rate is a good improvement, but not the industry's best, which doesn't make it defective.

Pictures below:
1- 1DX crop of banding and generally unusable noise levels at 5 stop push +100 shadows.
2- 5D4 equivalent (worst example I could create just now)
3- Same as above, but with some noise reduction to see how it handle the banding (color and smoothness)
4- The 5D4 image slightly cropped, just to show that even with a 5 stop push and 100 percent shadow recovery, it's usable.

 

[geekpower]

hi,
a test that you have to do to see the defect is a wide dynamic scene that includes deep shadow as well as bright areas. A real life situation. The quickest way to find such a scene is to find a dark room in your home with a window with some detail and bright sunshine outside. Expose for the scene outside and let the shadows in the room fall where they may . Then try some shadow lifting. If you don't get banding here, then it is not likely you will routinely find this kind of under-exposure in most things you will shoot.


iso 100
this:

what is this supposed to prove? the 2nd picture is not a push of the first, unless pushing 5 stops can magically turn a white pillow blue, and a smart phone into a laptop sleeve.

 

[privatebydesign]

hi,
a test that you have to do to see the defect is a wide dynamic scene that includes deep shadow as well as bright areas. A real life situation. The quickest way to find such a scene is to find a dark room in your home with a window with some detail and bright sunshine outside. Expose for the scene outside and let the shadows in the room fall where they may . Then try some shadow lifting. If you don't get banding here, then it is not likely you will routinely find this kind of under-exposure in most things you will shoot.


iso 100
this:

what is this supposed to prove? the 2nd picture is not a push of the first, unless pushing 5 stops can magically turn a white pillow blue, and a smart phone into a laptop sleeve.

Nice catch

Alternatively people could learn to use their cameras better, but no, wildly unrealistic expectations of 14 bit capabilities shouldn't get in the way of a rant, should it?

This from yesterday..........

 

[dcm]

hi,
a test that you have to do to see the defect is a wide dynamic scene that includes deep shadow as well as bright areas. A real life situation. The quickest way to find such a scene is to find a dark room in your home with a window with some detail and bright sunshine outside. Expose for the scene outside and let the shadows in the room fall where they may . Then try some shadow lifting. If you don't get banding here, then it is not likely you will routinely find this kind of under-exposure in most things you will shoot.


iso 100
this:

what is this supposed to prove? the 2nd picture is not a push of the first, unless pushing 5 stops can magically turn a white pillow blue, and a smart phone into a laptop sleeve.

Or move the plant? And open the door?

I'll assume jakdaniel1975 just picked the wrong first image when posting. But it would be nice to see an apples to apples comparison.

 

[tr573]

hi,
a test that you have to do to see the defect is a wide dynamic scene that includes deep shadow as well as bright areas. A real life situation. The quickest way to find such a scene is to find a dark room in your home with a window with some detail and bright sunshine outside. Expose for the scene outside and let the shadows in the room fall where they may . Then try some shadow lifting. If you don't get banding here, then it is not likely you will routinely find this kind of under-exposure in most things you will shoot.


iso 100
this:

what is this supposed to prove? the 2nd picture is not a push of the first, unless pushing 5 stops can magically turn a white pillow blue, and a smart phone into a laptop sleeve.

Nice catch

Alternatively people could learn to use their cameras better, but no, wildly unrealistic expectations of 14 bit capabilities shouldn't get in the way of a rant, should it?

This from yesterday..........

He posted the whole set of RAW's on DPR forums, and with both of those photos (the before and after which are of course different photos) if you do a reasonable conversion there's no banding.

It's only when you blast everything all the way to the right , which sure is a fun experiment, would not result in a nice exposure even if the banding weren't there. It's far too bright as anyone can see looking at it - you can dig a remarkable amount of detail out of the shadows, but you can't go kookoo on it.