Lens Parameters for Samyang 8mm on fullframe Canon 6d

[Frimann Bjornsson]

Hi there

I'm trying to get a nice stitch using my new Samyang  8mm fisheye on a Canon 6D fullframe.

So far I haven't been able to do a perfect stitch. 

I'm wondering if anyone cares to share their lens profile for this lens on a full frame body?

Or even share some points on how it's best to create a lens profile? 

Kind regards

Frímann

[Frimann Bjornsson]

One thought here. When taking a panorama that's supposed to be used for adjusting lens parameters, wouldn't it be best to take it outdoors, and using only control points for far away objects, to eliminate parallax point misalignment?

[UtahBob]

On Friday, January 22, 2016 at 10:12:04 AM UTC-5, Frimann Bjornsson wrote:

One thought here. When taking a panorama that's supposed to be used for adjusting lens parameters, wouldn't it be best to take it outdoors, and using only control points for far away objects, to eliminate parallax point misalignment? 

I'm using the Pro-Optic version of this 8mm lens.  I thought I had a good npp on my D200 with my 10.5 nikkor but one day I put it on my new D7000 and forgot to play with the npp.  To my surprise, where I had it set turned out to be the perfect place and the wall/ceiling join was impeccable.  So even though you think you are pretty good, there is always room to be better to the point the stitch is perfect.

With the D7000 and the 8mm, I used a method that John recommended of photographing an interior space with good floor, wall, and ceiling items for excellent control point coverage.  I used my shed that has shelves on all the walls, a tiled ceiling, and "stuff" all over the floor.  I shot about 10 spherical at f8 and 1.5m which seems to give me the best dof.  For each shot I moved the lower arm of my nodal ninja 3.  I then stitched all these including the nadir and picked the one that gave me the best result in terms of cp distance, etc.  I also looked at the sawtooth pattern that presents itself on the rotator of the nn3 when the npp is off.

I think this is the best method to come up with right lens parameters.  I'm not sure if you could use those of another user given that in my lens I have a pretty large shift parameter but it might be a starting place.

If you setup outside with no close objects, I think that will be problematic when inside but setting up inside should give you good outside stitches also.  At least that is what I'm seeing.

[Frimann Bjornsson]

From what I've heard this lens has the npp not as a point, but rather as a spherical (or at least bent) plane, so the location depends how far you are off the optical axis. So it matters where you measure your npp, based on where you will stitch the images. Say f.x. if I wan't to stitch 4 images per 360 degrees, that gives me 90° between images. Which means that I should set the npp +- 45° from the optical axis. 

It would be interesting to hear what are the best lens parameters that you have achieved with this lens (preferably on a full frame body). The best stitch I've done so far has the focal length set to 13.56mm, which is not even close to 8mm. I actually think that this lens is about 9.5mm, at least according to some tests I did in PTGUI.

Care to share your lens parameters?

[UtahBob]

On Friday, January 22, 2016 at 2:24:31 PM UTC-5, Frimann Bjornsson wrote:

There are sketches on the web somewhere that show angles for say 4 around or 6 around, etc. for fisheyes such as the 10.5 and your 8mm.  I don't know the links for those but have seen them posted here previously.  With the method John suggested, you don't need to care about setting up the npp with the 45 degrees, etc as the best final stitch of all the calibration panos will give you the right spot.

Here is the link: http://www.hugha.co.uk/NodalPoint/Index.htm#Samyang%208mm

I only have crop sensors so I didn't give you my numbers but here they are.  As you can see my FL is at 9.6

fl 9.597 

crop 1.529

hfov 144.5

180

a .05984

b -.02034

c .02633

d 1.254

e 19.97

My average cp distance on the calibration stitch was 1.1 with a low of .06 and a high of 4.9.  This is in a room with a close object of about 1 foot or so to around 25 feet on the  far end.  Seams were pristine.

[Frimann Kjerulf]

I didn't really get this method you talked about:

"With the D7000 and the 8mm, I used a method that John recommended of photographing an interior space with good floor, wall, and ceiling items for excellent control point coverage.  I used my shed that has shelves on all the walls, a tiled ceiling, and "stuff" all over the floor.  I shot about 10 spherical at f8 and 1.5m which seems to give me the best dof.  For each shot I moved the lower arm of my nodal ninja 3.  I then stitched all these including the nadir and picked the one that gave me the best result in terms of cp distance, etc.  I also looked at the sawtooth pattern that presents itself on the rotator of the nn3 when the npp is off."

Are you using this method to evaluate the position of the npp?

What is this sawtooth pattern you talk about?

ᐧ

Kær Kveðja / Kind Regards
Frímann Kjerúlf Björnsson
GSM: +354 822-3685

--
Many people are reading this forum via email and get every post delivered to their inbox. To limit the amount of data please do not add attachments or images to your posts; instead upload your files at a file sharing site (for example http://ge.tt/ ) and include a link in your message.
---
You received this message because you are subscribed to a topic in the Google Groups "PTGui Support" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/ptgui/-QTm6aQzly4/unsubscribe.
To unsubscribe from this group and all its topics, send an email to ptgui+un...@googlegroups.com.
To post to this group, send email to pt...@googlegroups.com.
Visit this group at https://groups.google.com/group/ptgui.
To view this discussion on the web visit https://groups.google.com/d/msgid/ptgui/0654bfee-4484-4989-9b33-5a6e4b6644f2%40googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

[UtahBob]

On Friday, January 22, 2016 at 4:49:38 PM UTC-5, Frimann Bjornsson wrote:

I didn't really get this method you talked about:

"With the D7000 and the 8mm, I used a method that John recommended of photographing an interior space with good floor, wall, and ceiling items for excellent control point coverage.  I used my shed that has shelves on all the walls, a tiled ceiling, and "stuff" all over the floor.  I shot about 10 spherical at f8 and 1.5m which seems to give me the best dof.  For each shot I moved the lower arm of my nodal ninja 3.  I then stitched all these including the nadir and picked the one that gave me the best result in terms of cp distance, etc.  I also looked at the sawtooth pattern that presents itself on the rotator of the nn3 when the npp is off."

Are you using this method to evaluate the position of the npp?

Yes, you are finding the optimum npp that will translate into the lens parameters that you can use going forward on new projects.  See the ptgui faqs 5.4.

The key to the method above is that you are shooting a spherical where you have control point coverage everywhere which allows the software to calculate the lens parameters.  If you use a normal room for instance you have the white walls and ceiling where there are no places for control points to be assigned and therefore the program isn't really evaluating the lens properly.  So a gym or anything with features all over will do.  Some people will create bullseye targets to tape on a blank wall and ceiling if they don't have an appropriate space so there will be places to assign cp.

By having the lens parameters as a starting point for future projects, the white wall and ceiling boundary will stitch seamlessly in the absence of control points for those areas.

In my calibration I started with a lower rail position that worked ok for me and them moved back and forth on the lower rail by half a mm and shot a pano at each location on the rail.  Then I stitched them all and picked the one that looked the best in terms of cp distance.

What is this sawtooth pattern you talk about?

 http://www.easypano.com/forum/display_topic_threads.asp?ForumID=1&TopicID=4162  and also item 3 on John's page ᐧhttp://www.johnhpanos.com/epcalib.htm  There is also a writeup somewhere in detail but I can't find it.

Some people don't like using the sawtooth but if I recall, my optimum npp in the calibration created a nadir with the least sawtooth.

Remember if you calibrate using 4 around, stick to that in future panos.  With full frame you might be able to go to 3?   I think if your seam locations are closer to the center of the lens - i.e. 4 around rather than 3, it is easier to evaluate visually where the cp are placed by the program since the further to the edge on the fisheye you get the more compression exists.  The 8mm Sam has a different lens config than say a Nikkor 10.5.  The Nikkor has a lot more compression near the edge and that seems to be why people like this 8mm so much, me included.

[Frimann Bjornsson]

Thanks for providing the parameters. I though I should be able to use your parameters, since the only difference between my setup and yours is the crop factor. 

Now I've spent a good two days trying to get this to work with the methods you recommended. Still I'm not able to align a proper blend.

I'd like to add that I have plenty of experience with PTGUI, when I was using my old Panasonic MFT camera, I regularly got perfect stitches with no errors.

I think I've got the NPP covered, at least to a decent enough precision. The sawtooth pattern is not visible as can be seen here: http://360.pan.is/samyangDemo/parallaxtest.jpg.

I've managed to get parallax errors down to a bare minimum. I aligned a toothpick about 20cm from the lens to houses at least 500m away. I'm shooting  4 + zenith + nadir, so images at 90°. Results can be seen here: http://360.pan.is/samyangDemo/parallaxtest.jpg The center picture is taken with the toothpick and houses exactly in the middle. The other four pictures were taken with the camera rotaded 45° to the left, right, up and down. As seen in the image there is not much parallax error visible, so I would say that I got that part covered. I realize that the parallax point of this lens shifts with the angle of incident, but since I'm taking images 90° apart, minimizing parallax errors at +/- 45°, where the stitch is most likely to be, should give me beset results.

Now to PTGUI, I can stitch the four horizontal pictures pretty well (average distance ~1, maximum distance ~14). But as soon as I introduce the zenith and nadir the alignment get's way off and average distance drops down to 5. This is the best I've been able to get so far:  http://360.pan.is/samyangDemo/ .  There are obvious errors, f.x. where the walls meet the ceiling. I would be greatly thankful if anyone could take a quick look at my project, and see if the can get a good fit, and tell me the secret! :)  My project can be downloaded here: http://360.pan.is/samyangDemo/ptgui.zip

The methood that I use in PTGUI was pretty solid when I was working with my Panasonic camera. It involved the following steps:

• Loading images
• Setting estimated lens parameters (or the one's from UtahBob)
• Placing the pictures in approximately correct locations around the sphere.
• Search for control points (delete bad points, and points on near objects)
• Optimize, with yaw, pitch and roll. With "field of view", "a", "b", "c", "horizontal" and "vertical shift" selected.
• Search again for control points (delete bad points, and points on near objects)
• Optimize again.

If anyone could help me on the right track here, or point me to what I might be doing wrong, that would be greatly appreciated!

Kind regards

Frímann

[Frimann Bjornsson]

Sorry, here comes the correct sawtooth link: http://360.pan.is/samyangDemo/sawtooth.png

[John Houghton]

Frimann, I got a good stitch (eventually):  http://ge.tt/56HWmHW2/v/0?c .  Why do you shoot at ISO3200?  You will get much better quality at ISO100, and control point assignment will be easier.  I think the nadir is probably upside down (is auto rotation set?) .  It certainly works better when rotated 180.  The zenith is peculiar. I don't know exactly what's wrong, but I assigned individual parameters and that did the trick. It might be upside down but rotating it didn't cure the problem on its own.

John

[UtahBob]

I tried the project and the closest I come is without the nadir which gets me an average cp of 3 but I have to viewpoint correct the zenith.  I'll add that I added each image one at a time and joined it to the next.  I can't really see a lot of parallax error in the source images but there is definitely some there between the zenith and the four around.  You can see this in the toothpick images for up and down which are off by a lot.  

It shouldn't be this hard.  You must be off a bit on the npp. I'm assuming your lens center is centered above the rotator?  I wonder if there is a full frame impact here somehow that I am not aware of shooting a crop sensor.

[Erik Krause]

Am 23.01.2016 um 18:51 schrieb Frimann Bjornsson:
> I think I've got the NPP covered, at least to a decent enough precision.
> The sawtooth pattern is not visible as can be seen here:
> http://360.pan.is/samyangDemo/parallaxtest.jpg.

Right-Left yes, Up-Down no. There is still a reasonable amount of parallax.

BTW.: You know that you loose half of your sensor pixels using the 8mm
on a full frame sensor? You'd get a far higher resolution on a crop
sensor: http://wiki.panotools.org/DSLR_spherical_resolution

--
Erik Krause
http://www.erik-krause.de

[John Houghton]

On Saturday, January 23, 2016 at 7:33:07 PM UTC, John Houghton wrote:

 I think the nadir is probably upside down (is auto rotation set?) .

Frimann, I see from the exif data that you shot raw images.  Can you check that the setting of the orientation tag in the metadata of all the images is the same (either all -90 or all +90) prior to converting to tiff/jpg?  Rotate any images that are out of step with the others.

John

[Frimann Bjornsson]

I actually just rotated the images that were in landscape to portrait using Preview (in Mac), that was just the nadir and zenith. Didn't think the rotation of the nadir and zenith mattered. Why is that btw?

Regarding the stitching, then I managed to get a much better parallax adjustment. After that I managed to get a pretty decent stitch, with just a few stitching artifacts. Minor ones though, which all disappeared when using SmartBlender.  So I'm a happy camper! 

Frímann

[Frimann Bjornsson]

I don't seem to be able to download that zip. Now I'm working with a better npp adjustment, and the stitching is going much better. Good point about the iso. 

On Saturday, January 23, 2016 at 7:33:07 PM UTC, John Houghton wrote:

[John Houghton]

On Sunday, January 24, 2016 at 12:51:02 AM UTC, Frimann Bjornsson wrote:

I actually just rotated the images that were in landscape to portrait using Preview (in Mac), that was just the nadir and zenith. Didn't think the rotation of the nadir and zenith mattered. Why is that btw?

It's important to get the orientation correct to ensure that the lens shift corrections are consistently applied to all of the images.  These parameters correct for the optical axis of the lens not being accurately aligned with the centre of the image frame (or crop).  If an image is upside down, the corrections are applied in the opposite direction to what they should be, making things worse rather than better.

I've filed a copy of my zip file at another site:  https://www.sendspace.com/file/oavffu .  Note that the project file assumes rotated nadir and zenith images, which were tiff. Included is an unedited panorama with no stitching errors, I think.

John

[Frimann Kjerulf]

Yes of course, that makes sense! Probably best to turn off auto rotation in-camera to be safe. 

ᐧ

Kær Kveðja / Kind Regards
Frímann Kjerúlf Björnsson
GSM: +354 822-3685

--

Many people are reading this forum via email and get every post delivered to their inbox. To limit the amount of data please do not add attachments or images to your posts; instead upload your files at a file sharing site (for example http://ge.tt/ ) and include a link in your message.
---
You received this message because you are subscribed to a topic in the Google Groups "PTGui Support" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/ptgui/-QTm6aQzly4/unsubscribe.
To unsubscribe from this group and all its topics, send an email to ptgui+un...@googlegroups.com.
To post to this group, send email to pt...@googlegroups.com.
Visit this group at https://groups.google.com/group/ptgui.

To view this discussion on the web visit https://groups.google.com/d/msgid/ptgui/3e057460-9d8c-4531-8f58-eed1916d2c29%40googlegroups.com.

[UtahBob]

On Sunday, January 24, 2016 at 3:38:51 AM UTC-5, John Houghton wrote:

It's important to get the orientation correct to ensure that the lens shift corrections are consistently applied to all of the images.  These parameters correct for the optical axis of the lens not being accurately aligned with the centre of the image frame (or crop).  If an image is upside down, the corrections are applied in the opposite direction to what they should be, making things worse rather than better.

John

That makes a lot of sense.  I guess it is really important for fisheyes since in my case the shift value is rather high for both my 8mm and 10.5mm.  I've always had orientation turned off so I never experience the problems associated.

I did see what I thought was an error on the picture frame that one seam cuts through.  I'd expect that he won't see it if he smart blends as it will route the seam to the blank wall. 

[John Houghton]

On Sunday, January 24, 2016 at 2:43:03 PM UTC, UtahBob wrote:

I did see what I thought was an error on the picture frame that one seam cuts through. 

Your eyes don't deceive you - there is a small error.  Running "Generate control points here" in that general area, which was not generously endowed with points, proved to be an effective cure.

John