Improving CHDK ND filter support

c_joerg

reyalp
If it works this way, you should be able to find this point just by adjusting the Tv value, starting with one that gives you Av < 8.

And probably with a hysteresis for in and outâ€¦

I want to understand some basics about NDâ€¦
How does the cam made exposure measure in AV mode when I set AV=8.0?
My understanding was, the cam opens aperture, do the exposure measure and closes only aperture while shooting.

What happen on S110 if I switch the ND in canon UI on? I see that the screen just goes a little bit dark. So the ND is really in? What happens if I do half press? Does the cam measure exposure (and focus) with ND in? Why the cam should do this? It made measure and focus much harderâ€¦.

What happens on SX50 after half press? Does ND really goes in or only at full shoot?

reyalp

c_joerg
I want to understand some basics about NDâ€¦
How does the cam made exposure measure in AV mode when I set AV=8.0?
My understanding was, the cam opens aperture, do the exposure measure and closes only aperture while shooting

On g7x, I can see the aperture opens and then goes to the set value on half press. How far it's open in normal live view depends on lighting. The ND on elph130 behaves similarly: If you point it at a very bright scene, the ND goes in just in live view, without half pressing. In darker conditions it might still go in on half press.

On g7x, the ND state changes when it's set in the UI, and from sound I don't think it goes out for auto exposure in half press. Canon manual ND has been quite primitive, IIRC only in g7x mk2 they added the option for auto modes to control it.

reyalp

ND related eventprocs etc (from elph130, but similar exist on many cameras)

PutInNdFilter
PutOutNdFilter
What we use for ND control

ReloadNdFilterAdjustment
implies there is some kind of camera specific ND calibration, though it could be actuator rather than value related.

TurnOffNdFilter
TurnOffNdFilterWithTiming
TurnOnNdFilter
TurnOnNdFilterWithTiming
Appear to be lower level functions than PutIn / PutOut

SetNdDriveVoltageSetting
GetNdMoveTime
SetNdMoveTime

task_Nd
Reverse engineering and hooking this might provide insight (and possible control, though it appears to be lower level than the event procs)

Possibly ND related LogCameraEvent strings
M:ND_OFF_E
M:ND_OFF_S
I don't see these in c_joerg's logs, or in logs from elph130 when ND is triggered.

c_joerg

reyalp
If it's using ND, it seems like there should be a light level (or Tv value) where the ND goes in and the Av drops below 8, like
Av = 8, no ND
move to 1 stop brighter light
Av = 4 (- 2 stops), ND in (+3-ish stops)

This confuses me more and moreâ€¦
I looks like that the sound which I here could be also the apertureâ€¦.

Make a test video. Had to do amplifier Audio to +20dB
All done in MF Mode
First Part: AV mode f5.6. If I do a half press. The sound which I here must come from aperture or?
Second part: TV Mode 1/10s. If I put my hands in front of the cam, I here same noise as from the first part. Not sure if this is aperture or NDâ€¦.

reyalp

Hard to know much based on sound unless you have a way to verify that you can distinguish between the aperture and the ND. On the G7X, I can tell they are different because putting in the ND manually in the UI makes a distinct sound.

You could try to do this on one of your cams by triggering the ND with CHDK. One way to do this in chdkptp would be something like

=press'shoot_half' repeat sleep(10) until get_shooting() sleep(1000) set_nd_filter(1) sleep(1000) set_nd_filter(2) sleep(1000)

Normally, when you call set_nd_filter, it just sets a flag to tell the camera to put in the ND in the next half press. If you call it inside half press, it takes effect immediately. The above code waits 1 second and then puts in the ND, which should allow you to distinguish the sound. You should also see the live view get darker.

edit:
Oops, had the ND control values reversed. Fixed.

c_joerg

reyalp
Hard to know much based on sound unless you have a way to verify that you can distinguish between the aperture and the ND. On the G7X, I can tell they are different because putting in the ND manually in the UI makes a distinct sound.

Same on S110 or G1x. Sound between aperture and the ND is different.
Hard to say on SX50 which noise is aperture or ND.

The first part of the video shows you this execution in chdkptp

=press'shoot_half' repeat sleep(10) until get_shooting() sleep(3000) set_nd_filter(1) sleep(3000) set_nd_filter(2) sleep(3000)

I changed only sleep to 3s from your code. You can see that ND goes in and out. But you see here also three different brightness steps (off,in,out).

In the second part I tried to change the brightness slowly until I hear a sound. But at this time the brightness did not changed on live view. So not sure what that means. May be the cam compensates the ND step immediately when ND goes on and off?

reyalp

c_joerg
I changed only sleep to 3s from your code. You can see that ND goes in and out. But you see here also three different brightness steps (off,in,out).

This is weird. In half press, I'd expect other exposure params to stay fixed, but it seems like the camera must be changing something else on in or out.

In the second part I tried to change the brightness slowly until I hear a sound. But at this time the brightness did not changed on live view. So not sure what that means. May be the cam compensates the ND step immediately when ND goes on and off?

I think this is normal when the camera puts ND on in live view. It's the same on my ND only cameras like elph130, firmware knows the ND value and adjusts the live view exposure at the same time. I can hear the click just aiming from indoors to a sunny window and back, and there's no iris to confuse things.

c_joerg

reyalp
I think this is normal when the camera puts ND on in live view. It's the same on my ND only cameras like elph130, firmware knows the ND value and adjusts the live view exposure at the same time.

But that is different on S110 and G1x . If I switch ND on, the live view goes just dark and got normalized after 1s.

reyalp

c_joerg
But that is different on S110 and G1x . If I switch ND on, the live view goes just dark and got normalized after 1s.

Right, what I'm saying is that automatic ND behaves the same on ND-only ixus/elph style cameras and "hidden" ND cameras like sx50.

I suspect manual ND is done the way it is to give users a visual cue, but it could just be simpler implementation.

c_joerg

reyalp On both elph130 and d10, I noticed that the zoom level seems to change very slightly with/without ND.

Same on G1xâ€¦.
The Video shows ND switch in the middle of the video. The video has only 40 pictures with 10fps. The video is also a 1920*1080 crop from full picture.

Another interesting think is changing of noise in the video when the cam goes from ISO1000 to ISO130. I donâ€™t use any noise reduction

I tried to correct this Video
With

align_image_stack -c 24 -i -x -y -a aligned_ *.jpg

But result looks worthâ€¦

The log file includes the full run. ND switch is close to the end.

rawopint.csv

207.22KB

lapser

c_joerg

reyalp On both elph130 and d10, I noticed that the zoom level seems to change very slightly with/without ND.

Same on G1xâ€¦.
Another interesting think is changing of noise in the video when the cam goes from ISO1000 to ISO130. I donâ€™t use any noise reduction

Adding a filter moves the focal point back a small amount, depending on the thickness of the filter.
http://www.lcas-astronomy.org/articles/display.php?filename=filters_and_focus_shift

I've noticed a slight magnification change, and slight blurring, in my time lapses before and after an ND filter switch. It's not very noticeable to most people, so I haven't done anything to fix it. I suppose the way to correct it would be to change the focus after an ND switch, but I'm not sure it's worth the hassle.

I have two G1X cameras. I usually set the focus at the hyperfocal point, which works with one camera. But the other camera gets too blurry with the ND filter in, so I set that camera to infinity focus.

The way to correct your time lapse video would be to zoom slightly in or out after an ND filter switch.

Here's an example video:

Watch the tree on the right after the Sun goes down in the above time lapse. The filter switches out around 40 seconds, which moves the tree slightly, and changes the sky pattern a little. Nobody has ever noticed it (except me).

Regarding the ISO change, I switched the filter in at 1/500 sec 100 ISO in this video. I switched back out at 1/34.6 sec 100 ISO, which makes the first shot without the filter 1/250 sec 100 ISO. Note the ISO is always at minimum if you set your switch point to a fast enough shutter speed.

Here's the part of my log file for this video that shows the ND filter switch:

Shot# dEV sEV Shutter  ISO   Bv* IntD PD mVÂ°F  >50  >90 H-M1 Hist   M#1  IMG_
 1115  -4 274  1/34.6  100  615*   51  3 7849    0    0   38   -3Â«  -42Â« 9821
Auto OUT (274) 273Â« -1
 1116  -1  -1~  1/250  100  614    53  8   80    0    0   38   -5Â«  -43Â« 9822
 1117  -4   0   1/248  100  615    49  1 7859    0    0   38   -3Â«  -42Â« 9823

reyalp

c_joerg

reyalp On both elph130 and d10, I noticed that the zoom level seems to change very slightly with/without ND.

Same on G1xâ€¦.

One way to deal with this would be to use log to determine when the ND changed and crop / scale the images to match. Wouldn't address the focus change lapser mentioned though.

lapser

reyalp
One way to deal with this would be to use log to determine when the ND changed and crop / scale the images to match. Wouldn't address the focus change lapser mentioned though.

There's usually enough depth of field to handle the small focus change from the ND filter. I had a problem when I got too fancy and set the focus at the hyperfocal distance computed by CHDK. That's as close as you can set the focus and still have infinity in acceptable focus.

But when you switch the ND filter from out to in, the focus moves in even further, causing the blur. The solution was to focus on infinity in the first place, and not use the hyperfocal point.

You can see the blurring when the filter switches in at 1:16 in this sunrise time lapse (starting at 1:06)

https://youtu.be/fZ7V9Wy2UN4?t=66

c_joerg

lapser
The way to correct your time lapse video would be to zoom slightly in or out after an ND filter switch.

This is probably the easiest wayâ€¦

I have tried a lot of stabilization algorithms; strangely they all fail in such simple situations.

lapser
Nobody has ever noticed it (except me).

I've noticed that sometimes. I just thought it was a gust of wind. 😉
The more you deal with the subject, the more you pay attention.

What you can also see very well at the log file is that the ND value from the props works very well.

lapser

c_joerg

lapser
The way to correct your time lapse video would be to zoom slightly in or out after an ND filter switch.

This is probably the easiest wayâ€¦
I have tried a lot of stabilization algorithms; strangely they all fail in such simple situations.

lapser
Nobody has ever noticed it (except me).

I've noticed that sometimes. I just thought it was a gust of wind.
The more you deal with the subject, the more you pay attention.

I tried the zoom method with this time lapse video from last Wednesday. I ended up cropping out 4 pixels horizontal, and 3 vertical from the clip with the filter in (4K video). I did it in the rendering progam since it's hard to crop accurately with Lightroom (I'll keep working on it).

Anyway, it seems to have worked pretty well. The ND filter goes out at 40 seconds into the video:
It's hard to see anything happening. If you watch closely, you'll see a slight brightness glitch in the sky.

The video has a lot of brightness fluttering with the Sun in the frame. This could be my 4 million exposures shutter wearing out. But I also mistakenly saved the photos as jpg instead of raw, so it could relate to the camera creating jpgs with blown pixels in the frame.

One thing I've noticed is that the ND filter density needed varies with the brightness you set for the filter to switch in or out. On one of my G1X cameras, the measured density is around 18 ev96 greater when switching in than when switching out.

To deal with this, I save the brightness before the switch, and measure the brightness in the first photo after the switch. I adjust the next photo so its brightness matches the saved brightness before the switch. What this means is that I have a one photo "flash" after a switch, which I can correct for in Lightroom. I can then correct the ND filter density parameter for the next time lapse if necessary.

c_joerg

lapser
I tried the zoom method with this time lapse video from last Wednesday. I ended up cropping out 4 pixels horizontal, and 3 vertical from the clip with the filter in (4K video). I did it in the rendering progam since it's hard to crop accurately with Lightroom (I'll keep working on it).

Yes, I know, unfortunately in Lightroom you canâ€™t crop in pixel.

lapser
Anyway, it seems to have worked pretty well.

It works greatâ€¦

lapser
One thing I've noticed is that the ND filter density needed varies with the brightness you set for the filter to switch in or out. On one of my G1X cameras, the measured density is around 18 ev96 greater when switching in than when switching out.

Thatâ€™s interestingâ€¦.

lapser
To deal with this, I save the brightness before the switch, and measure the brightness in the first photo after the switch. I adjust the next photo so its brightness matches the saved brightness before the switch. What this means is that I have a one photo "flash" after a switch, which I can correct for in Lightroom. I can then correct the ND filter density parameter for the next time lapse if necessary.

Maybe I would do this in postprocessingâ€¦
My plan is to write a tool which does the metering on RAW data files, like rawoplib but only on the CR2 data. One advantage about this, I can select an area where I do the metering. Then I would modify the â€˜Exposure2012â€™ in the Lightroom XMPâ€™s similar as I descript here:
https://chdk.setepontos.com/d/12790

reyalp

lapser
One thing I've noticed is that the ND filter density needed varies with the brightness you set for the filter to switch in or out. On one of my G1X cameras, the measured density is around 18 ev96 greater when switching in than when switching out.

This is strange. There's no way the actual ND value can with brightness, so I'd guess there's a bug somewhere or a hardware side effect like the aperture moving when the ND state changes.

lapser

reyalp

lapser
One thing I've noticed is that the ND filter density needed varies with the brightness you set for the filter to switch in or out. On one of my G1X cameras, the measured density is around 18 ev96 greater when switching in than when switching out.

This is strange. There's no way the actual ND value can with brightness, so I'd guess there's a bug somewhere or a hardware side effect like the aperture moving when the ND state changes.

The 18 ev96 difference only happens on the camera I sent in to Canon to be refurbished. They replaced the entire lens assembly. I suspect it's a small calibration error. The focus in the refurbished camera is also a little off compared to my other G1X.

The ND switch out occurs at 1 ev darker than the switch in (what you call hysterisis I think). So it sounds like the conversion from ev96 to shutter speed may be a little off at different shutter speeds. I've also found that the measured ev96 of an ND switch is a little different at different switching brightness, in both cameras.

"In theory, theory and practice are the same. In practice, they're not!" _ Anonymous
"Reality is the murder of a beautiful theory by a gang of ugly facts." _ JohnClonts

reyalp

I thought sx710 had a "hidden" ND, because PutInNdFilter made a noise and a big difference in exposure.

However, running ndtest.lua, I happened to notice the DOF was a lot deeper for the ND-in shots. So I tried using Av mode with aperture set to F/8 (rather than max Av), and ndtest only reported an ND value of ~117 instead of ~305 as in the previous runs. When I tested Av overrides, there was detectable change up to around F/11, which would roughly match the F/8 + "ND" value.

Looking at the ROM code, sx710hs PutInNdFilter appears to be roughly
MoveIrisToAv(GetUsableMinAv())
😛 🤣 :lol

My impression is sx260hs (another "hidden" ND camera) code is the same. The code on G7x (which has an ND in the Canon UI) is totally different.

I would be interested if anyone can confirm the DOF changing significantly with the ND "in" on sx260hs or other supposed "hidden" cameras. If the "hidden" ND is not actually a thing, ND support in CHDK will be simplified considerably...

edit:
Additional data point: The non-hidden ND cameras have asserts for "NdActuator.c" and an ND task. These do not appear to be present in supposed "hidden" ND cams like SX260, SX50 and SX710

edit:
I mixed up Min and Max a couple times
GetUsableMinAv returns the smallest aperture (e.g. F/11), GetUsableMaxAv returns the largest (e.g F/3.2)
The propcase we call PROPCASE_AV_MIN returns a value like "max" 😳

srsa_4c

reyalpI would be interested if anyone can confirm the DOF changing significantly with the ND "in" on sx260hs or other supposed "hidden" cameras.

I can, I made such an experiment some years ago, with my sx280.
This year, however, I noticed dirt on my "ND in" test shots (and even during half shoot with ND in), which made me believe that the ND does exist in some form...
... until I applied curves on my non-ND test images.
See attached pictures, most spots are there on both shots (F8, full zoom, focus on infinite, "ND" in & out).

« Previous Page Next Page »