Fujifilm GFX100 II: IBIS Bug? Reader Comments, Plausible Explanation but Unanswered Question
Related: digital sensor, Dr S, Electronic First Curtain Shutter, Electronic Shutter, exposure, Fujifilm, Fujifilm GFX100 II, Fujifilm medium format, Fujifilm mirrorless, image stabilization, In-Body Image Stabilization, Jim S, medium format, Other World Computing, reader comment, sensor readout transit time, shutter, Stephen S
re: Fujifilm GFX100 II: Terrible Blur with Image Stabilization (IBIS) at Relatively High Shutter Speeds
We know that electronic shutter (ES) has issues including image deformation which cannot be avoided other than by reducing the sensor readout transit time with each generation of camera, something Sony and Nikon have mastered, being 15X to 20X faster than Fujifilm medium format.
Single-shot 16-bit captures: Fujifilm GFX 100 II/100S/100 all take ~1/3 second and ~1/6 second for 14-bit. Why is it so much slower than 35mm cameras? After all, the Sony A7R V is 60 megapixels but can get the job done in 1/200 sec or so.
The core issue is that the Fujifilm GFX100 II appears to have more IBIS issues than its GFX100S/100 predecessors, based on field shooting (see images below, taken from dozens of problem images). Extreme blur appears to be new behavior versus using electronic shutter (ES) almost exclusively for five (5) years on the GFX100 and GFX100S.
However, I cannot rule out that some particular set of shooting conditions produced such atrocious results. But what puzzles me is how I could shoot with ES + IBIS for five years and not see such consistently poor outcomes. That is what makes me think that the GFX100S II has something problemetic at work, not the fact that it has the usual limitations of ES.
While one explanation seems highly plausible (see comments by Christoph K), why did the GFX100S seemingly not manifest these issues?
As yet I have not done A/B tests and I do not have a good answer. I need similar conditions to what I had that day in order to do a decent A/B of real field conditions (vs trying to move the camera myself).
Sensor readout transit time
Jim Kasson is the real expert on sensor readout transit time. See his posts on the matter, starting with More on GFX 100 II electronic shutter speeds.
In the previous post, I reported the disappointing finding that, in spite of Fujifilm’s claims to increased readout speed, the scan time of the electronic shutter in 14-bit and 16-bit single shot modes was essentially the same as the GFX 100S and the GFX 100.
...So there is a mode that yields faster readout rates than the GFX 100S, but you have to set the camera up right to get it, and it comes at a cost.
Unchanged* sensor readout time aside, why does it appear that the GFX100 II performs so badly vs the GFX100S/100? Or is it perhaps just conditions that I never noticed for five (5) years?
* For single-shot mode in both 14-bit and 16-bit capture mode, the CH (continuous high) mode goes to 12-bit as per Jim Kasson’s ahgove.
Christopher K writes:
The read out time for a Sony A7R V / A7R IV / Leica M11 is about 100 ms at 14 bit (and about 50 ms at 12 bit), but not 1/200 s = 5 ms. That’s Sony A1 territory.
So, at 14 bit A7IV/V 100 ms GFX 100/100s/100 II 150 ms sounds very reasonable for an equivalent sensor technology
DIGLLOYD: my assumptions were wrong it seems, based on using the A1 for a good long time. But it still puzzles me why the GFX100 II showed such bad results. There might still be some other behavior at work, in addition.
Stephen S writes:
Lloyd, I could hardly believe your images but it took only a couple of minutes to confirm the problem with my GFX 100 II set to electronic shutter. In my limited testing it is worse with IS set to shooting only, better with a lens with OIS. Strangely with IS continuous and a shutter speed ~1/focal-length there can be eight or ten entirely satisfactory images but then one with blur along the bottom edge.
I've used the GFX 100 since it was released, almost always with electronic shutter; I skipped the GFX 100s. I don't recall ever seeing anything like this with the GFX 100; certainly if it occurred with regularity I'd have noticed, and I do make some use of non-OIS lenses. Makes one wonder if only the cameras with "improved" IBIS are affected, I don't know. I'm keeping the older camera and will do some experiments with it.
I've never seen anything like this with the Sony A7R IV and Sony A7R V, which I usually use with electronic shutter
I suppose I'll be using EF full time except when on a heavy tripod with no wind. I'm always well stopped-down, so the bokeh impact of EFCS with fast lenses is not going to be a problem.
I'm puzzled, disappointed, and annoyed.
...following up on prior comment above: When I began to study this I was getting frequent blurred images resembling yours, once every several exposures. In some of these as in your examples, the blurring was really too severe to be from angular movement of the camera associated with hand-holding, so it must have been spurious movement of the sensor attributable to the IBIS system. The more I tested the more the blur, when present, settled down to the lower edge of the image which is the end, not the beginning of the exposure. You'd expect the opposite with hand-excited motion.
And strangely, the more I tested, the rarer the blurred images became. Today, in more than forty images with the GF 50mm and a Canon 135mm TS/E, with varying shutter speeds around and below 1/focal-length, no blur at all in any image. I'm avoiding lenses with OIS because turning OIS off turns off IBIS.
I do not understand this!
Can it be that the camera's IBIS system, once it has some exercise, settles down to work correctly? That's just too crazy!
DIGLLOYD: rightly so. Not in five years have I had such actively destructive results. I am sure I could do better in all cases with IBIS disabled. And these are very solid shutter speeds eg 1/60 to 1/100 second for a 55mm (48mm equivalent) lens, hardly pushing the envelope!
With bugs, there can be strange runs of behavior. Curiously, I have a CFExpress card reader whose behavior has the same “streak” of fail/work but always failing if I plug in my iPhone 15 Pro Max first. Stuff just sometimes happens that way. At this point, I suppose I have to cross test the GFX100S vs GFX100 II, but of course I cannot reproduce a motion scenario consistently so it might take a lot of shots to diagnose. Or maybe a few—depending.
Dr S writes:
Before testing and finding an alignment issue and before I left the camera store, in the desert I popped on the 50mm lens and took a few images. I thought I had moved the camera while shooting a couple but after reading your blog entry this morning I now know it is the camera.
There was sufficient shutter speed to take an image. You will notice some are fine and some are atrocious. Until something is fixed I wonder what the workaround is. All OOC jpegs. Even one is oof when no movement is seen. Hmm?
...Yes. I just got notification it has been sent back from repair. When I checked its status, the serial number of the body returned is different from the one I sent. My assumption is they sent a new body.... I hope so. It is scheduled to arrive tomorrow Fedex overnight. I'll do another test......tomorrow when I get the shipment. I'll let you know. BTW have you tested its eye-AF? When they talk about surgical AF, I wonder if they are referring to its ability to focus on the eye. For Landscape images I will still check with MF. I am looking forward for using the EVF as long as the new body works properly.
...Got the replacement body and tested ES with and without IS. Not only OOF but vertical lines are wavy as in the previous email link. I hope this is an issue that can be fixed via FW. EF seems fine with cursory testing. I haven't used much slower SS. My 100S does not exhibit this behavior. Now I will mount the 20-35 to determine whether or not the previous body was defective or whether or not I will scream.
...Just read your comment and the readers comments. I need to wrap my head around how I am going to take any image based upon the issue. In 2023, we shouldn't have to. Again i will test the 100s tomorrow. If it is bright outside and I know the ss will be fairly brief..... 1/300s or shorter. I don't think EF will be that terrible. However, I would like to use ES all the time. Maybe Nikon will come out with a larger sensor cam at some point and apply their ES engineering like they have on the Z9 and Z8.
DIGLLOYD: Dr S had a peculiar sharpness issue with the GFX100 II which now seems to be well explained by this IBIS issue.
Jim S writes:
As I understand it, IBIS from every manufacturer is designed to compensate for camera movement based on the assumption that the entire sensor is exposed at the same time (true global shutter). Most of the time, this assumption doesn’t cause a problem due to the amount of camera movement and sensor scan speed errors not exceeding the resolving power of the lens/camera combination. But with a long sensor scan speed and extremely high resolution, it’s like trying capture the image in a fifth dimension: length, width, height, time + time-shift. The sensor scan speed serves as a second shutter, which spreads a relatively short amount of time for the exposure over a longer amount of time required to scan the sensor. To picture this, imagine the sensor scan speed was a whopping five hours. The exposure would stay the same at say 1/60 sec.; however, each pixel would serve as a virtual time lapse frame, and your first exposed pixel and your last exposed pixel represent different points in time, for example, the first pixel recorded at noon, the last recorded at 5pm.
That said, I’m a bit shocked by the level of blur in the image you showed. Typically IBIS + sensor scan speed blur is something you can’t see unless at least viewing at 50% magnification. What you’re showing is massive blur. I have never seen this on the 100S nor 50SII. What this looks like to me is what would happen if I started moving the camera to set up the next panoramic frame before the exposure was completed from the previous frame. And I’m assuming the sensor would have to be scanned in two horizontal halves at the same time to produce the two bands of blur shown in your example.
Perhaps better explained by saying: the slow sensor scan speed can cause a form of rolling shutter for which IBIS wasn’t designed to compensate
On the 50mp GFX sensor, the sensor scan in ES was quite loud, so I could listen and wait to move the camera until it stopped. Can’t remember if the 100mp sensor’s ES scan is audible.
...[followup] Yeah, it has to be a firmware bug with IBIS specific to the 100 II. The blurred images you posted look as if IBIS quits working and is moving the sensor back to the zero/locked position before the sensor scan in ES is completed.
DIGLLOYD: agreed, the explanation (see next comment) makes sense, but why have I not ever seen such bad behavior with GFX100S? An A/B in similar windy conditions is what I need to compare the two.
Sure looks like an IBIS bug to me too. But... maybe it's just a limitation? Again, why not the GFX100S/100 showing such issues. Maybe over 5 years the conditions just never happened?
Christoph K writes:
Honestly, I was wondering a bit the same thing like you. And I am not 100 % sure how the sensor read out exactly works. I might ask Jim at Dpreview about it.
But my guess is: If we use MS, the shutter (slot) let’s say takes about 1/200 s to pass the sensor. Then it’s closed (I think the amount of light of every pixel is translated to a voltage which is stored in tiny capacitors) and it does not matter that it takes longer (1/3 - 1/6 s) to read out the sensor (= discharge the capacitors), there is no more light gathered on the sensor during that period of time. Therefore IBIS only has to control the camera movement during 1/200 s in this example.
It seams to be similar with EFC (open) plus MS (close) but I am not totally sure. But with ES its clear the the sensor gathers light for 1/3 s at 16 bit resp. for 1/6 s at 14 bit, independently of the exposure time. Enclosed an example of ES with the GFX 50s where the person is moving during the 1/4 s ES time, the camera was still. I mean here we would not expect IBIS to compensate for the movememt of the object but it shows that a camera movement during 1/4 s could be tremendous and impossible to compensate by IBIS.
DIGLLOYD: I agree on the times. AFAIK, the GFX100 II still takes about 1/3 secondto readout the sensor for 16-bit captures (sensor readout transit time).
And I think the basic premise is correct: with mechanical shutter, the exposure time equates to the time that IBIS must maintain the sensor in a fixed/stable position for the image eg a 1/80 second exposure IBIS need only stabilize for 1/80 second.
However, my assumption is that ES necessarily resets groups of rows as it prepares to scan them (otherwise the top vs bottom would have wildly different exposures 1/80- second at bottom vs 1/80 + 1/3 second at top). That cannot be. Thus my assumption is that IBIS operates in groups of rows (“bands”), as the sensor readout is done electronically, clearing a band of sensor rows prior to reading it, while waiting the requisite time before reading it. Or something like that.
My assumption has been that IBIS would operate continually during that time. But that means that IBIS has to keep that sensor stable for a full ~1/3 of a second, thus for a 1/80 second requiring IBIS sensor stabilization 26X as long. Thus producing bands of sharp/unsharp.
Fujifilm GFX100 II + Fujifilm GF 35-70mm f/4.5-5.6 WR @ 42.6mm equiv (51.8mm) + polarizer Breakthrough Photography X4
ENV: Mt Conness, altitude 12200 ft / 3719 m, 60°F / 15°C
RAW: vignetting corrected
[low-res image for bot]
Fujifilm GFX100 II + Fujifilm GF 35-70mm f/4.5-5.6 WR @ 57.6mm equiv (70mm) + polarizer Breakthrough Photography X4
ENV: Mt Conness, altitude 12200 ft / 3719 m, 60°F / 15°C
RAW: vignetting corrected
[low-res image for bot]
Fujifilm GFX100 II + Fujifilm GF 35-70mm f/4.5-5.6 WR @ 28.8mm equiv (35mm) + polarizer Breakthrough Photography X4
ENV: Mt Conness, altitude 11600 ft / 3536 m, 60°F / 15°C
RAW: vignetting corrected
[low-res image for bot]
