Re: Fwd: Sony a7R1 and a7R3 introduce a pixel shift in each consecutive image capture
Khoroshyy Petro <[email protected]> Mon, 29 May 2023 09:44:48 +0200
| Newsgroups | gmane.comp.multimedia.gphoto.user |
|---|---|
| Message-ID | <CAEoO5zpPK1BZ2Ned94+VH_rsxY1s7AXD8kWPAp6gj_rTJ7K4-w@mail.gmail.com> |
--===============5559757872589723934== Content-Type: multipart/alternative; boundary="00000000000088f0e705fcd04377" --00000000000088f0e705fcd04377 Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: quoted-printable Just an idea. Does this camera work with an old M42 lens? Try to put in some old lens in it. That would locate if problem is in the lens or in the camera. Best. Petro. On Mon, 29 May 2023, 09:38 Rafal Mantiuk, <[email protected]> wrote: > Just an update on the issue of the pixel shift on Sony cameras. > > We retook the images (100 images) of a checkerboard for three > different cameras, registered the position of the crossings and > plotted them over time. The results: > > https://www.dropbox.com/s/uhyllcwkn9joo1l/Canon_T1i.png?dl=3D0 > https://www.dropbox.com/s/e73thcjiiebqqus/Sony_a7R1.png?dl=3D0 > https://www.dropbox.com/s/e73thcjiiebqqus/Sony_a7R1.png?dl=3D0 > > x-axis is time > y-axis is the relative pixel shift, for different crossings on the > checkerboard. It is a fractional number as the checkerboard detection > works on subpixels. > > This confirmed our previous observation: no shift for Canon and quite > a bit of a shift for Sony. > > I am not sure whether there is a way to either prevent this shift or > determine what this shift is. Sony software can control this shift > (for the pixel-shift super-resolution capture), so I assume there must > be a way to do that. I would appreciate any information or input on > that. > > For now, I hope that this post is a precaution for those who want to > use Sony cameras for computer vision. > > Rafal > > > On Fri, Mar 31, 2023 at 12:42=E2=80=AFPM Rafal Mantiuk <[email protected]= > wrote: > > > > Thank you for all the suggestions. > > > > a7r3 indeed has the pixel shift feature (which is off) and in-sensor > > image stabilization, but a7r1 does not but it shows the same problem. > > > > I measured multiple times with the same and different lenses and with > > both cameras. I measured at most 30 frames, I can easily measure more > > (and report back). Surprisingly, the motion seems to be always the > > same and not random. > > > > I did not move the camera or checkerboard as it is difficult to get > > accurate and controllable motion. > > > > Rafal > > > > > > On Fri, Mar 31, 2023 at 11:41=E2=80=AFAM Khoroshyy Petro <khoroshyy@gma= il.com> > wrote: > > > > > > Maybe you can try longer sequence? > > > > > > Does it happen always it only measured once? Target moving? Camera > moving? > > > > > > > > > > > > > > > > > > On Fri, 31 Mar 2023, 12:31 Rafal Mantiuk, <[email protected]> wrote: > > >> > > >> The Sony cameras (a7r1 and a7r2) do not have a mirror. The lens has = a > > >> mechanical aperture, but I doubt it can cause vibrations. > > >> I would expect a more random movement if mechanical vibration was a > cause. > > >> > > >> Canon T1 does have a mirror, but there are no pixel shifts between > the shots. > > >> > > >> Thanks, > > >> > > >> Rafal > > >> > > >> On Fri, Mar 31, 2023 at 11:26=E2=80=AFAM Khoroshyy Petro <khoroshyy@= gmail.com> > wrote: > > >> > > > >> > Hi. > > >> > Does the mirror vibration move the camera? > > >> > Petro. > > >> > > > >> > On Fri, 31 Mar 2023 at 11:25, Rafal Mantiuk <[email protected]> > wrote: > > >> >> > > >> >> Hi, > > >> >> > > >> >> When capturing a sequence of images using > > >> >> --capture-image-and-download, each image approx 12 seconds apart > from > > >> >> the previous one, the pixels in the consecutive images are > shifted, as > > >> >> can be seen in this GIF: > > >> >> > > >> >> > https://www.dropbox.com/s/91llepa8hwpm81u/sony_a7r3_2.8-90_v2.gif?dl=3D0 > > >> >> > > >> >> The GIF shows 10 frames. The shift is the most obvious when > > >> >> transitioning from the last to the first frame. > > >> >> > > >> >> Another example, 30 frames for a different lens: > > >> >> > > >> >> > https://www.dropbox.com/s/9pa0aoxy3rs5dwc/sony_a7r1_1.8-50_jpeg_v3.gif?dl= =3D0 > > >> >> > > >> >> This is a big bummer when trying to use the camera for computer > > >> >> vision, calibration or photogrammetry. > > >> >> > > >> >> I cannot observe such a problem with Canon (Rebel T1). > > >> >> > > >> >> Image stabilization is switched off - both optical and in-camera. > a7R1 > > >> >> should not have in-camera stabilization or a movable sensor. I > tried > > >> >> also the lens without optical image stabilization. The same happe= ns > > >> >> for RAW and JPEG capture. > > >> >> > > >> >> Interestingly, the shift is quite predictable - the image always > goes > > >> >> up and right. This looks more like a "feature" rather than a bug. > > >> >> > > >> >> Does anyone have any idea what could be causing this pixel shift? > > >> >> > > >> >> Thanks, > > >> >> > > >> >> Rafal > > >> >> > > >> >> > > >> >> _______________________________________________ > > >> >> Gphoto-user mailing list > > >> >> [email protected] > > >> >> https://lists.sourceforge.net/lists/listinfo/gphoto-user > > > _______________________________________________ > Gphoto-user mailing list > [email protected] > https://lists.sourceforge.net/lists/listinfo/gphoto-user > --00000000000088f0e705fcd04377 Content-Type: text/html; charset="UTF-8" Content-Transfer-Encoding: quoted-printable <div dir=3D"auto"><div>Just an idea.<div dir=3D"auto">Does this camera work= with an old=C2=A0 M42 lens?</div><div dir=3D"auto">Try to put in some old = lens in it. That would locate if problem is in the lens or in the camera.</= div><div dir=3D"auto">Best.</div><div dir=3D"auto">Petro.</div><div dir=3D"= auto"><br></div><div dir=3D"auto"><br></div><br><br><div class=3D"gmail_quo= te"><div dir=3D"ltr" class=3D"gmail_attr">On Mon, 29 May 2023, 09:38 Rafal = Mantiuk, <<a href=3D"mailto:[email protected]">[email protected]</a>>= wrote:<br></div><blockquote class=3D"gmail_quote" style=3D"margin:0 0 0 .8= ex;border-left:1px #ccc solid;padding-left:1ex">Just an update on the issue= of the pixel shift on Sony cameras.<br> <br> We retook the images (100 images) of a checkerboard for three<br> different cameras, registered the position of the crossings and<br> plotted them over time. The results:<br> <br> <a href=3D"https://www.dropbox.com/s/uhyllcwkn9joo1l/Canon_T1i.png?dl=3D0" = rel=3D"noreferrer noreferrer" target=3D"_blank">https://www.dropbox.com/s/u= hyllcwkn9joo1l/Canon_T1i.png?dl=3D0</a><br> <a href=3D"https://www.dropbox.com/s/e73thcjiiebqqus/Sony_a7R1.png?dl=3D0" = rel=3D"noreferrer noreferrer" target=3D"_blank">https://www.dropbox.com/s/e= 73thcjiiebqqus/Sony_a7R1.png?dl=3D0</a><br> <a href=3D"https://www.dropbox.com/s/e73thcjiiebqqus/Sony_a7R1.png?dl=3D0" = rel=3D"noreferrer noreferrer" target=3D"_blank">https://www.dropbox.com/s/e= 73thcjiiebqqus/Sony_a7R1.png?dl=3D0</a><br> <br> x-axis is time<br> y-axis is the relative pixel shift, for different crossings on the<br> checkerboard. It is a fractional number as the checkerboard detection<br> works on subpixels.<br> <br> This confirmed our previous observation: no shift for Canon and quite<br> a bit of a shift for Sony.<br> <br> I am not sure whether there is a way to either prevent this shift or<br> determine what this shift is. Sony software can control this shift<br> (for the pixel-shift super-resolution capture), so I assume there must<br> be a way to do that. I would appreciate any information or input on<br> that.<br> <br> For now, I hope that this post is a precaution for those who want to<br> use Sony cameras for computer vision.<br> <br> Rafal<br> <br> <br> On Fri, Mar 31, 2023 at 12:42=E2=80=AFPM Rafal Mantiuk <<a href=3D"mailt= o:[email protected]" target=3D"_blank" rel=3D"noreferrer">[email protected]= </a>> wrote:<br> ><br> > Thank you for all the suggestions.<br> ><br> > a7r3 indeed has the pixel shift feature (which is off) and in-sensor<b= r> > image stabilization, but a7r1 does not but it shows the same problem.<= br> ><br> > I measured multiple times with the same and different lenses and with<= br> > both cameras. I measured at most 30 frames, I can easily measure more<= br> > (and report back). Surprisingly, the motion seems to be always the<br> > same and not random.<br> ><br> > I did not move the camera or checkerboard as it is difficult to get<br= > > accurate and controllable motion.<br> ><br> > Rafal<br> ><br> ><br> > On Fri, Mar 31, 2023 at 11:41=E2=80=AFAM Khoroshyy Petro <<a href= =3D"mailto:[email protected]" target=3D"_blank" rel=3D"noreferrer">khoros= [email protected]</a>> wrote:<br> > ><br> > > Maybe you can try longer sequence?<br> > ><br> > > Does it happen always it only measured once? Target moving? Camer= a moving?<br> > ><br> > ><br> > ><br> > ><br> > ><br> > > On Fri, 31 Mar 2023, 12:31 Rafal Mantiuk, <<a href=3D"mailto:m= [email protected]" target=3D"_blank" rel=3D"noreferrer">[email protected]</a= >> wrote:<br> > >><br> > >> The Sony cameras (a7r1 and a7r2) do not have a mirror. The le= ns has a<br> > >> mechanical aperture, but I doubt it can cause vibrations.<br> > >> I would expect a more random movement if mechanical vibration= was a cause.<br> > >><br> > >> Canon T1 does have a mirror, but there are no pixel shifts be= tween the shots.<br> > >><br> > >> Thanks,<br> > >><br> > >> Rafal<br> > >><br> > >> On Fri, Mar 31, 2023 at 11:26=E2=80=AFAM Khoroshyy Petro <= <a href=3D"mailto:[email protected]" target=3D"_blank" rel=3D"noreferrer"= >[email protected]</a>> wrote:<br> > >> ><br> > >> > Hi.<br> > >> > Does the mirror vibration move the camera?<br> > >> > Petro.<br> > >> ><br> > >> > On Fri, 31 Mar 2023 at 11:25, Rafal Mantiuk <<a href= =3D"mailto:[email protected]" target=3D"_blank" rel=3D"noreferrer">mantiuk@= gmail.com</a>> wrote:<br> > >> >><br> > >> >> Hi,<br> > >> >><br> > >> >> When capturing a sequence of images using<br> > >> >> --capture-image-and-download, each image approx 12 s= econds apart from<br> > >> >> the previous one, the pixels in the consecutive imag= es are shifted, as<br> > >> >> can be seen in this GIF:<br> > >> >><br> > >> >> <a href=3D"https://www.dropbox.com/s/91llepa8hwpm81u= /sony_a7r3_2.8-90_v2.gif?dl=3D0" rel=3D"noreferrer noreferrer" target=3D"_b= lank">https://www.dropbox.com/s/91llepa8hwpm81u/sony_a7r3_2.8-90_v2.gif?dl= =3D0</a><br> > >> >><br> > >> >> The GIF shows 10 frames. The shift is the most obvio= us when<br> > >> >> transitioning from the last to the first frame.<br> > >> >><br> > >> >> Another example, 30 frames for a different lens:<br> > >> >><br> > >> >> <a href=3D"https://www.dropbox.com/s/9pa0aoxy3rs5dwc= /sony_a7r1_1.8-50_jpeg_v3.gif?dl=3D0" rel=3D"noreferrer noreferrer" target= =3D"_blank">https://www.dropbox.com/s/9pa0aoxy3rs5dwc/sony_a7r1_1.8-50_jpeg= _v3.gif?dl=3D0</a><br> > >> >><br> > >> >> This is a big bummer when trying to use the camera f= or computer<br> > >> >> vision, calibration or photogrammetry.<br> > >> >><br> > >> >> I cannot observe such a problem with Canon (Rebel T1= ).<br> > >> >><br> > >> >> Image stabilization is switched off - both optical a= nd in-camera. a7R1<br> > >> >> should not have in-camera stabilization or a movable= sensor. I tried<br> > >> >> also the lens without optical image stabilization. T= he same happens<br> > >> >> for RAW and JPEG capture.<br> > >> >><br> > >> >> Interestingly, the shift is quite predictable - the = image always goes<br> > >> >> up and right. This looks more like a "feature&q= uot; rather than a bug.<br> > >> >><br> > >> >> Does anyone have any idea what could be causing this= pixel shift?<br> > >> >><br> > >> >> Thanks,<br> > >> >><br> > >> >> Rafal<br> > >> >><br> > >> >><br> > >> >> _______________________________________________<br> > >> >> Gphoto-user mailing list<br> > >> >> <a href=3D"mailto:[email protected]"= target=3D"_blank" rel=3D"noreferrer">[email protected]</a>= <br> > >> >> <a href=3D"https://lists.sourceforge.net/lists/listi= nfo/gphoto-user" rel=3D"noreferrer noreferrer" target=3D"_blank">https://li= sts.sourceforge.net/lists/listinfo/gphoto-user</a><br> <br> <br> _______________________________________________<br> Gphoto-user mailing list<br> <a href=3D"mailto:[email protected]" target=3D"_blank" rel= =3D"noreferrer">[email protected]</a><br> <a href=3D"https://lists.sourceforge.net/lists/listinfo/gphoto-user" rel=3D= "noreferrer noreferrer" target=3D"_blank">https://lists.sourceforge.net/lis= ts/listinfo/gphoto-user</a><br> </blockquote></div></div></div> --00000000000088f0e705fcd04377-- --===============5559757872589723934== Content-Type: text/plain; charset="us-ascii" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit Content-Disposition: inline --===============5559757872589723934== Content-Type: text/plain; charset="us-ascii" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit Content-Disposition: inline _______________________________________________ Gphoto-user mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/gphoto-user --===============5559757872589723934==--