NN3 + Canon 400D + 15mm Fiseye Settings??? [Archive] - Nodal Ninja Forum

PDA

View Full Version : NN3 + Canon 400D + 15mm Fiseye Settings???



gunsuka
09-07-2011, 09:12 PM
I bought my NN3 in 2008, tried it a few times but was never happy with the setup.

I recently upgraded it by adding the RD3L 6-8-30 Rotator EZ Leveler II (http://store.nodalninja.com/products/NN3-MKII-RD3L-6%252d8%252d30-Rotator-EZ-Leveler-II.html) (VERY nice product that one is).

I'm trying the setup with a Canon 400D and the Canon 15mm Fisheye, have spent days trying to get the alignment right. I'm close, but just not close enough for perfection every time.

The guide at johnhpanos.com (http://www.johnhpanos.com/epcalib.htm) is good but I still feel I don't have it quite right.

Searching these forms I've seen others trying to figure out the setup (http://www.nodalninja.com/forum/showthread.php?3251-Canon-400D-XTi-Canon-15mm-(fisheyes)) of this camera/lens combo but never any answer.

Anyone have the numbers for this camera/lens combo?

John Houghton
09-07-2011, 11:56 PM
I really don't understand how you can take days trying to get the alignment right. You can set up the panohead in about 10 minutes and get excellent stitching. If you are judging the success of the setup by the quality of the stitched results, then it may be that your stitching process is not as good as it should be. OTOH, if you test the setup by contriving situations where any parallax effects will be made visible and measurable, (as in the sticky tape on the window tests in the tutorial), then you will be in a position to tweak the setup to reduce any such effects to a minimum if not eliminate them altogether. Remember that a fisheye lens does not have a single no parallax point, so absolute perfection is not to be expected. Nevertheless, excellent stitching can usually still be obtained even when the setup is out by a mm or two.

If you have got the setup as good as you can get it, shoot a full 360x180 panorama, including a zenith + two nadir shots taken with the head rotated 180 degrees (of yaw) between them. Include something quite close to the camera (like a pole) in the middle of one of the horizontal image overlaps. If the resulting images don't stitch well enough, there should be enough telltale clues available to diagnose why. If you can let me have copies of the images I will be happy to investigate.

John

hindenhaag
09-08-2011, 12:42 AM
Hi,

If you do not find settings for you equipment in the NN database, you can use the wiki one:

http://wiki.panotools.org/Entrance_Pupil_Database

For LRS = Lower rail setting you have to measure or use the data you can find. You have a "offset" from camera body to the reference point of the vertical rail to basis of CP-2 camera plate which may be different for NN models. For NN3 MKII you have to add 13mm to H1 without added NN Nadir Adapter, with it 7mm. Check this with Smooth method.

http://www.easypano.com/forum/display_topic_threads.asp?ForumID=1&TopicID=4162

URS = upper rail setting You get it with H1 + H2 = 39 + 45 = 84.

You have to test these settings with you own equipment with John's methods and for LRS for the best with the method of Smooth. Because you stitch the shots around PTGui will correct extra differences between lenses and camera bodies like minimal sensor displacements for example.

http://michel.thoby.free.fr/Fisheye_history_short/Beyond-the-pupil.html This will explain what John told you about there is no "exact no parallax point".

Regards,
Heinz

gunsuka
09-08-2011, 07:24 AM
10 minutes, I wish. I've tried on and off for years. I blew another 5 hours on it tonight alone.

I can get it close, but never perfect. I've resorted to making 1mm changes at a time and trying again but still just can't get it perfect.

http://dl.dropbox.com/u/12473059/TokyoPano/alignment.2011.9.8.jpg

John Houghton
09-08-2011, 08:42 AM
It doesn't have to be perfect. That looks fine to me, but it isn't a complete test for assessing the setup of your panohead. Please shoot a panorama exactly as suggested in the last paragraph of my previous response. If you make the images available (half size jpeg at medium quality is sufficient), then we can see if the setup is good enough. Choose a suitable venue, though. The last place to use is a small room with white walls and ceiling - i.e. you want somewhere rich in features for assigning control points on.

John

hindenhaag
09-08-2011, 08:53 AM
Hi, your teeth are cutting to the left, so move left on the lower rail. MM by MM. Till they turn and cut to the right, then move backwards to the right. LRS should be found in between. Sometimes you even have to move by 0.5mm.

Then check URS. I use a reference point i front of a board. When you turn right and the reference moves with this direction, you have to move forward. When it turns against it, you have to go backwards. Think of two people walking side by side. When you follow your guy, your going forward = camera forward. When he turns around and walks away from you, you have to go backwards to reach him again. Camera backwards. To be sure, you have to check the pictures on the PC. For the last check, you have to use an overlay of the second pic to the first one and click layers on and off in PS. Reference should not move at the right setting. Even using live view, this is only a "Pre Set". When you look through your viewfinder, you will often think the reference is moving. So looking through it, often gives a wrong result. With my Nikon I use Transfer2 to import and use ViewNX2 to pre check the pics. You can click from one to the other and zoom in if needed. I do not know your Canon software. This often avoids Photoshop overlays.

300302301305304303306307308

http://web.me.com/hindenhaag/filechute/Searching%20NPP.zip You can download the pics to see them better. I personally take center to the right and to the left.

May be right now you can get an imagination.

Regards,
Heinz

John Houghton
09-08-2011, 11:30 AM
Heinz, Your series of images presumably were taken to check for parallax but they should ideally be taken at a yaw increment of 60 degrees, if that is what you are using for an actual panorama (6 around). The yaw increment for your test images appears to be considerably smaller than this.

John

hindenhaag
09-08-2011, 12:13 PM
John, You are right. The yaw increment is smaller. I added these pics just to give an example of what I was trying to explain.

Regards,
Heinz

badders
09-09-2011, 12:24 AM
I can get it close, but never perfect. I've resorted to making 1mm changes at a time and trying again but still just can't get it perfect.
What is your definition of "perfect"? Do you mean the alignment of the shots as you look down onto the nadir? Or do you (as you should) mean what a fully stitched panorama looks like? Because that is the correct way to check if your pano head is calibrated. If you can shoot a panorama where there are no stitching errors then it matters not a jot what the view down onto the nadir looks like.

John Houghton
09-09-2011, 02:02 AM
What is your definition of "perfect"? Do you mean the alignment of the shots as you look down onto the nadir? Or do you (as you should) mean what a fully stitched panorama looks like? Because that is the correct way to check if your pano head is calibrated. If you can shoot a panorama where there are no stitching errors then it matters not a jot what the view down onto the nadir looks like.
Exactly, though with the caveat that you need to shoot and stitch the test panorama intelligently. One panorama with no stitching errors is no guarantee that all panoramas will stitch without errors.

John

gunsuka
09-10-2011, 01:42 AM
I tried a pano of a large room on Friday night using the NN configured as it was when I did the snapshot above. It was bad :-(

I continued to mess with the upper and lower arm. Moving the upper arm had the biggest impact, it is about 98% now I would say.

I did the following pano today:

http://gunther.ca/pano/deck/

John Houghton
09-10-2011, 04:29 AM
That's a good result, but it's difficult to account for the minor stitching errors without having copies of the camera images. It does need levelling, BTW. See http://www.johnhpanos.com/levtut.htm .

John

gunsuka
09-18-2011, 12:33 AM
Ok, I was back at it again this weekend. I did another pano (http://www.gunther.ca/pano/coastguard/) and it came out OK.

I know the alignment still is not stitch perfect though, so I took another set in daylight with some poles as suggested in this thread. I put the JPG images in an archive, they can be download here (http://dl.dropbox.com/u/12473059/TokyoPano/AlignPoles.rar) if anyone wants to take a look and can possible tell me what is out of alignment.


Robert

hindenhaag
09-18-2011, 03:09 AM
Robert,

Looks quite good. Re check the nadir with some stitching areas.

To check totally please shoot 2 rows at +/- 35 plus zenith and Nadir shots for daylight set. Looks good as well till now.

http://www.vrwave.com/panoramic-lens-database/canon/

Heinz

Sorry, missed the second row downloading.

John Houghton
09-18-2011, 06:54 AM
The images stitched well for me and gave a panorama without any stitching errors,aside from some small misalignments in the overhead wires, which might be due to the wires moving about in wind. Masking out the offending bits cured this blemish. I put my project file at http://www.johnhpanos.com/poles-jh.zip . I don't see any near poles for checking parallax, though. The nadir shots show the LRS to be ok, but there's a small alignment error that might be corrected by twisting the camera slightly on its mount to the upper rail, if that's possible. It's not serious. You can see the resultant small parallax shift of the panohead base relative to the floor in this animation:

http://www.johnhpanos.com/poles.gif

But generally, the setup looks to be quite satisfactory.

John

hindenhaag
09-18-2011, 08:27 AM
I got a similar result. Small stitching errors on the wires in one area and a small one which occurred on one rail on top besides the tree. In first case I found errors in the nadir with rests of tripod showing. So I had to mask pics 1860 1863 1866 besides the nadir shots.

322323324

Remember to save John's Stitch as a template to get good results in the future.

Heinz

hindenhaag
09-18-2011, 08:57 AM
del

gunsuka
09-18-2011, 10:15 PM
Remember to save John's Stitch as a template to get good results in the future.

Thanks for the feedback guys, I am using PTGUI and it stitched without any problems (often it says it can't find the points to connect photos, but not this time). However my finished stitch has problems along the fence that I noticed, they were quite obvious. I don't have access to them at the moment but will post them later, maybe I just need to update PTGUI or change some setting.


Heinz, what do you mean save John's stitch as a template? I can use it somehow to import photos like a workflow or something?

hindenhaag
09-18-2011, 11:20 PM
Template: http://www.panoguide.com/forums/qna/4299/

Yes you should upgrade to PTGui Pro and then get the latest Beta version which offers a much better spread for fisheye lenses for example, masking function etc etc.

Heinz

John Houghton
09-18-2011, 11:51 PM
Template: http://www.panoguide.com/forums/qna/4299/

Robert, Also, don't overlook the Help facility in PTGui. F1 will take you straight to help for the current screen. Then select Main Menu in the left panel and scroll down to Apply Template and click the link to Working with templates.

Your stitching errors on the fence can be caused by any of the following:

- badly positioned control points and absence of points (the optimizer cannot directly align areas where there are no points)

- inaccurate lens parameters

- parallax errors due to inadequate setup of the panohead

- movement of objects in the scene
.
Innacurate lens parameters arise from the optimizer's attempts to align the control points. If there are badly positioned control points or parallax etc, the evaluated lens parameters are likely to be wrong, leading to imperfect correction of lens distortions and inaccurate warping of the images. See http://wiki.panotools.org/Optimization (http://wiki.panotools.org/Optimization) for further details.
.
John

hindenhaag
09-19-2011, 12:38 AM
Good morning John,

"I don't see any near poles for checking parallax, though."

I think you might have to explain what one has to do and how to place a near reference point in the "overlap area" to fine tune the settings in the stitched pano to check for parallax errors caused by wrong settings. Would be a help for more unexperienced users.

Thx in advance,
Heinz

John Houghton
09-19-2011, 04:00 AM
I think you might have to explain what one has to do and how to place a near reference point in the "overlap area" to fine tune the settings in the stitched pano to check for parallax errors caused by wrong settings. Would be a help for more unexperienced users.

Heinz, Good morning. I already said this in an earlier post in this thread:
"If you have got the setup as good as you can get it, shoot a full 360x180 panorama, including a zenith + two nadir shots taken with the head rotated 180 degrees (of yaw) between them. Include something quite close to the camera (like a pole) in the middle of one of the horizontal image overlaps. If the resulting images don't stitch well enough, there should be enough telltale clues available to diagnose why. If you can let me have copies of the images I will be happy to investigate."
.
Robert has already seen my full tutorial on the subject at http://www.johnhpanos.com/epcalib.htm . But to clarify: one needs to have an object close to the camera (50-100cm, say - distance not critical) such that it will appear in two neighbouring shots where they overlap. Behind the near object should be a distant background with features that act as a reference to make any parallax shift of the object obvious and measurable. The object does not have to be a pole: it can be the edge of a door, a piece of sticky tape on a window, a railing, etc. Naturally, you need to avoid control points placed on the near object - have points only on the far background.
.
Of course, the checks for parallax can be performed just using the Liveview feature on the camera (if available), but actual photos need to be shot as for a panorama for the evidence to be seen here.

John