Announcement

Collapse
No announcement yet.

RS-1 Is this the wrong way round, if so...

Collapse
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • #31
    Hi John.
    Here are the separate images and the ptgui project file.
    https://drive.google.com/open?id=0B9...Fd2RTFkd2FtUzA

    I am pretty familiar with Photoshop, but have never used it for 360 photos, and have no idea at all how photoshop would help me correct stitching errors. That's probably something I need to research and learn about, perhaps on Youtube or something, though I had not imagined having to tweak each image so much once the NN gear is correctly configured.

    If there is something that is obviously wrong with my panos, I would love to know, the stitching is very obvious in tight spaces...

    Thanks again again good sir,

    Mark.

    Originally posted by John Houghton View Post
    Mark. It's difficult to diagnose the cause of stitching problems just by looking at the stitched panorama. Stitching errors can be due to many factors and once again I must request copies of the images in order to understand better what's going on. Shooting in tight spaces can be challenging. The problems usually occur towards the zenith and nadir because of the changing position of the entrance pupil, but errors there are usually covered up by the zenith and nadir shots (straight up and down). In any event, minor stitching errors can usually be corrected by editing in Photoshop fairly easily. That's often quicker than battling to get a perfect stitch straight out of PTGui, whatever the cause of the stitching errors.

    John

    Comment


    • #32
      Mark, I wasn't able to do a lot better than you with my alignment. Only by applying some masking was I able to get rid of most of the errors. There is some (small) parallax visible between images in the horizontal row, but whether correcting that would make much difference I don't know. My project file is at https://www.sendspace.com/file/hs0165 . Also, using the alternative intelligent blender Smartblend produces an almost perfect stitch without resorting to masks.

      John

      Comment


      • #33
        John, that was certainly better than my effort. Did you achieve that just with some masks or did you add some points too?

        I will try that Smartblend, it seems PTGUI's website is down at the moment, I may try find it elsewhere too.

        Thanks John, time for bed...

        Originally posted by John Houghton View Post
        Mark, I wasn't able to do a lot better than you with my alignment. Only by applying some masking was I able to get rid of most of the errors. There is some (small) parallax visible between images in the horizontal row, but whether correcting that would make much difference I don't know. My project file is at https://www.sendspace.com/file/hs0165 . Also, using the alternative intelligent blender Smartblend produces an almost perfect stitch without resorting to masks.

        Comment


        • #34
          Mark, Yes I did add more control points. With my usual panorama editing skills, i finally got this result:https://www.sendspace.com/file/f3sdx4 . I think you should concentrate on getting the horizontal parallax error down to as near zero as you can manage. The attached image shows the magnitude of the error, which is quite significant for such close objects. You should do tests with an object at about 50cm away viewed against a background as far away as possible (eg a distant scene outside a window).

          John

          Comment


          • #35
            Hi John,

            Sorry, I only just saw this response.
            BTW, Smartblend works really well, almost always better than PTGUI for the blending...

            Presumably the test for this is the adjustment of the lens ring slider, not the horizontal bar setting, which we tested (sawtooth cd test) and found to be correct at 5.9cm.
            So I would put a pin close to the lens (3 or 4 inches away), line it up with a distant object, and pan the camera from side to side, making sure the pin remains in the exact same spot, right?

            I also tried this test with TILTING the camera, not panning... but I suppose since my motion will be panning rather than tilting, I should focus on that?

            Thanks, almost there I feel...
            Mark.

            Originally posted by John Houghton View Post
            Mark, Yes I did add more control points. With my usual panorama editing skills, i finally got this result:https://www.sendspace.com/file/f3sdx4 . I think you should concentrate on getting the horizontal parallax error down to as near zero as you can manage. The attached image shows the magnitude of the error, which is quite significant for such close objects. You should do tests with an object at about 50cm away viewed against a background as far away as possible (eg a distant scene outside a window).

            John

            Comment


            • #36
              Mark, Generally speaking, you should test for horizontal parallax with the same setup you would use for shooting a panorama. The focus setting usually affects the entrance pupil position, so set the focus as for a panorama. Also, when you pan the camera from side to side, the near object should appear to be in the same position relative to the distant background at the two consecutive yaw positions: eg at -30 and +30 if shooting 6 shots around. What happens at other angles is not relevant. At 3 or 4 inches away, a pin may well be out of focus. You actually did the required test accidentally in your panorama, where the near object was a vertical rail on the staircase banister. See the epcalib tutorial section 4.

              John

              Comment


              • #37
                Hi John,
                Yep, I have done quite a lot of testing today, and my findings are that the settings on the NN head are way to crude to reliably set and re-create... a setting of 28.5 mm was to far one way, and 28.8mm (my guess) was too far the other way! So we are talking about the correct position being somewhere within that THIRD OF A MILLIMETER! That is crazy, and with a tightening screw which moves the ring on the bracket as you tighten it, it is almost impossible to find the right point, let alone match it precisely on location. I have it very close now... somewhere between those two measurements.

                I have attached three photoshop PSD files which have two layers, one where the camera is panned right and one where it is panned left, and toggling the top layer on and off it is easy to see the offset. The (so called) 2.86 one is the best, I guessed that measurement because it looked like it was a tiny fraction over 28.5mm...

                The stitching I have done with that best setting is pretty good, very small noticeable fault...

                I wonder if the settings are so crude because it is a ring mounted NN device, or does that not make a difference?

                Cheers, and John, thanks again again.

                Mark.


                Originally posted by John Houghton View Post
                Mark, Generally speaking, you should test for horizontal parallax with the same setup you would use for shooting a panorama. The focus setting usually affects the entrance pupil position, so set the focus as for a panorama. Also, when you pan the camera from side to side, the near object should appear to be in the same position relative to the distant background at the two consecutive yaw positions: eg at -30 and +30 if shooting 6 shots around. What happens at other angles is not relevant. At 3 or 4 inches away, a pin may well be out of focus. You actually did the required test accidentally in your panorama, where the near object was a vertical rail on the staircase banister. See the epcalib tutorial section 4.

                John
                Attached Files

                Comment


                • #38
                  Mark, Well you seem to have ended up with a satisfactory setting. It's so much easier to judge the alignment of the pairs of images by using PTGui to align them using a template to set the lens parameters and then you only need 3 or 4 control points on the background. Then output a layered rectilinear psd file for viewing in Photoshop or just use the Detail Viewer, switching the images off/on via the Include Images list on the Create Panorama tab.

                  John

                  Comment


                  • #39
                    Hi (again) John/Nick.

                    I am tweaking my system, and have something that confused me. On my bracket, the horizontal adjustment takes the camera closer or further from the rotation axis point. The only other adjustment I have is to adjust how far fowards or backwards the camera is in relation to the axis point. When the camera is pointing downwards, there is no adjustment for moving the camera forwards and backwards. That same adjustment will instead be moving the camera up and down, ie closer and further from the ground. SO. when I am shooting directly down, there is no adjustment to move the camera in such a way as to remove this stitching error you see here attached. Click image for larger version

Name:	rotate 180 ok far ng close.jpg
Views:	2
Size:	99.1 KB
ID:	28638
                    The stitching error in the bubble is a problem when things are close (like the car interiors I am trying to do) but not when the subject is 1m away or more. What it seems to me is that the bracket is not centered, ie, when you rotate the bracket 180, the bubble should be in precisely the same spot. The two adjustments I have available to me will make no difference to this horizontal split, and will only take the two halves of the bubble closer or further from each other.

                    I hope I am making sense here, but I really want to have no stitching problems, and I cannot nail it with very close subjects.

                    Could the bracket be at fault??
                    tx
                    Mark.

                    Originally posted by John Houghton View Post
                    Mark, From the appearance of the nadir, it looks like the lower rail setting isn't optimum. See the tutorial referenced in the earlier post.

                    John

                    Comment


                    • #40
                      Mark, Your photo indicates that the bottom rail setting needs adjusting to shift the camera in towards the bubble level by 1-2 mm. The misalignment in the direction at right angles is very small and even if you cannot do anything about it, should not be a problem. Remember that you cannot eliminate parallax completely for fisheye lenses because the entrance pupil is not in a fixed position, so just do the best you can.

                      John

                      Comment


                      • #41
                        Hi John,
                        I did adjust the horizontal bar, and found it did improve, (though it makes no sense to me as the skewing is not explained... the bubble "should" get longer and thinner when adjusting the horizontal bar, but there should be no skewing?)

                        But also interesting something shown in the attached animated gif.
                        Click image for larger version

Name:	left-right-up-down.gif
Views:	2
Size:	263.9 KB
ID:	28645



                        If you compare the UP DOWN frames, you see the lens is too far forwards
                        and when you see the LEFT RIGHT frames, the lens appears to be too far back

                        is that assessment right? Though the LEFT RIGHT will also be affected by the horizontal bar setting, which is correct now... but the up down should not be affected at all by the horizontal bar setting as there is no movement on the X axis.

                        If that is so, then surely it would follow that I should set the lens ring bracket adjustment to favour the UP DOWN reduction of parallax, and to use the horizontal bar to adjust for the LEFT RIGHT panning?

                        thanks for your expertise John!
                        Mark.

                        Originally posted by John Houghton View Post
                        Mark, Your photo indicates that the bottom rail setting needs adjusting to shift the camera in towards the bubble level by 1-2 mm. The misalignment in the direction at right angles is very small and even if you cannot do anything about it, should not be a problem. Remember that you cannot eliminate parallax completely for fisheye lenses because the entrance pupil is not in a fixed position, so just do the best you can.John
                        Last edited by markpalmos; 12-04-2016, 04:26 AM.

                        Comment


                        • #42
                          Mark, Your assessment is correct, but the horizontal bar adjustment should not be used to correct panning. As I have said, you cannot eliminate all parallax effects with a fisheye lens and you will just have to compromise. As well as optical considerations, there are physical aspects of the setup that can conspire to degrade results - such as flexing of bars and tripod legs, and soft materials on which the tripod might be stood. Try to minimize parallax in the directions where you have things close to the camera. You can maybe exercise some control over where the seams are positioned so that they don't cut across near features where glitches are likely to show.

                          John

                          Comment


                          • #43
                            Hi John,

                            I hear you, that fisheye has inherent issues... I am very close to "as good as possible" - so will leave it as it is now. I'd like to learn to do some Photoshop stitch fixes...

                            When you make corrections in Photoshop, presumably you are exporting all the shots as separate layers, then tweaking them by transforming (slightly distorting) the layers to fix minor stitching errors? Do you do this by viewing layers involved in the stitching error with a 50% transparency so you can visually fix errors? I have searched youtube but can't find a tutorial on that?

                            tx
                            Mark


                            Originally posted by John Houghton View Post
                            Mark. It's difficult to diagnose the cause of stitching problems just by looking at the stitched panorama. Stitching errors can be due to many factors and once again I must request copies of the images in order to understand better what's going on. Shooting in tight spaces can be challenging. The problems usually occur towards the zenith and nadir because of the changing position of the entrance pupil, but errors there are usually covered up by the zenith and nadir shots (straight up and down). In any event, minor stitching errors can usually be corrected by editing in Photoshop fairly easily. That's often quicker than battling to get a perfect stitch straight out of PTGui, whatever the cause of the stitching errors.John

                            Comment


                            • #44
                              Mark, Fixing glitches can be done in a variety of ways. You can often stop them occurring in the first place by using Smartblend, which can route seams around objects. Then, you can make minor corrections using copy / paste / shift-or-rotate / flatten. (See tutorial: http://www.johnhpanos.com/glitch_correction.html ).

                              Working on the nadir and zenith areas of the equirectangular image is not generally recommended because of the extreme distortion in those areas. Instead, either extract rectilinear views one way or another (eg. create cubic tiles) or remap the equirectangular image to bring both the zenith and nadir to the horizontal level where they are easy to retouch: load into PTGui the equirectangular image (lens type equirectangular, fov 360). Then run the numerical transform tool with y/p/r adjustments 90/0/90. Generate the output image for editing in Photoshop. Then input the edited equirectangular image back into PTGui and apply numerical transform parameters 270/-90/0 to flip the image back to its original orientation.

                              And yes, you can output separate layers or a layered PSD file and use layer masks to manually merge images or parts of images to eliminate stitching errors. There are loads of Photoshop tutorials on youtube for using the various transform tools and layer masks.

                              John

                              Comment


                              • #45
                                Thanks John,
                                All valuable advice. I will have a look at your tutorial first.
                                I'm pretty familiar with PS, but not with 360 photography. A search for photoshop ptgui on youtube came up with nothing useful. But I will get creative with search terms.
                                tx again John.
                                Mark.


                                Originally posted by John Houghton View Post
                                Mark, Fixing glitches can be done in a variety of ways. You can often stop them occurring in the first place by using Smartblend, which can route seams around objects. Then, you can make minor corrections using copy / paste / shift-or-rotate / flatten. (See tutorial: http://www.johnhpanos.com/glitch_correction.html ).

                                Working on the nadir and zenith areas of the equirectangular image is not generally recommended because of the extreme distortion in those areas. Instead, either extract rectilinear views one way or another (eg. create cubic tiles) or remap the equirectangular image to bring both the zenith and nadir to the horizontal level where they are easy to retouch: load into PTGui the equirectangular image (lens type equirectangular, fov 360). Then run the numerical transform tool with y/p/r adjustments 90/0/90. Generate the output image for editing in Photoshop. Then input the edited equirectangular image back into PTGui and apply numerical transform parameters 270/-90/0 to flip the image back to its original orientation.

                                And yes, you can output separate layers or a layered PSD file and use layer masks to manually merge images or parts of images to eliminate stitching errors. There are loads of Photoshop tutorials on youtube for using the various transform tools and layer masks.

                                John

                                Comment

                                Working...
                                X