Luminous Landscape Forum
Raw & Post Processing, Printing => Digital Image Processing => Topic started by: Jack Flesher on February 17, 2007, 10:29:51 am
-
With all the questions on CS3's abilities and since pictures are worth thousands of words, and with the recent stitching discussions, I thought I'd post an illustrated example of CS3's new stitching capabilities.
Here are four frames captured by just leveling the camera on my tripod and panning these four frames, with no particular attention paid to panning at the nodal point. I was using the mirror-up and 2-sec timer function on my 5D. I was not in any particular hurry, but even so the total elapsed time for all four captures was 12 seconds -- or in reality closer to 6 seconds net if I subtract the self-timer delays.
First off, I know this is not a great image, but one chosen to show how well CS3 automatically handles a difficult stitch. Note that I converted raw images to these small jpegs and the quality of the raw stitch is incredible, though difficult to show on the web with these small jpegs. Also note that I did NO tweaking to the images before or after the stitch and no adjustment of the final stitch or seams; these are merged "as-shot":
1)
(http://jack.cameraphile.org/albums/album05/panoframe_1.jpg)
2)
(http://jack.cameraphile.org/albums/album05/panoframe_2.jpg)
3)
(http://jack.cameraphile.org/albums/album05/panoframe_3.jpg)
4)
(http://jack.cameraphile.org/albums/album05/panoframe_4.jpg)
~~~
Here is the dialog that pops up when you open photomerge. I have added "open" files to the merge:
(http://jack.cameraphile.org/albums/album05/panoframe_dialog.jpg)
Here is the finished automatic cylindrical or "sphreical" projection:
(http://jack.cameraphile.org/albums/album05/Panoframe_spherical.jpg)
Here is the finished automatic perspective correct or "flat" projection:
(http://jack.cameraphile.org/albums/album05/panoframe_perspective_correct.jpg)
Here is a closer in crop of the seams -- frames 2 and 4 have the opacity dialed down to 80% so you can see the quality of the seam:
(http://jack.cameraphile.org/albums/album05/panoframe_layers.jpg)
Offered FWIW -- Cheers,
-
Offered FWIW -- Cheers,
[a href=\"index.php?act=findpost&pid=101375\"][{POST_SNAPBACK}][/a]
Also FWIW, I presented a seminar to 200 pros and students at a conference yesterday and, amongst the stuff I showed, I demoed the Auto-Align and Auto-Blend feature on two images shot like you did, hand held, but on a 17mm lens. Normally I would consider it too much work to stitch two such wide shots but CS3 did it perfectly, first time, and in front of 200 people.
There was a moment of silence and then everyone applauded!
Photomerge is a good thing.
-
That's very impressive, Jack. I've wasted hours trying to get straight lines straight and a neat join when stitching subjects like this taken from a close distance. It looks as though the cost of the upgrade to CS3 would be justified by this feature alone.
One problem that often occurs with stitched images is lack of a perfectly smooth tonal gradation in the sky, especially when the light source (sun) is to the far left or right of the scene, as in a dawn. How does PS3 handle this?
By the way, have you traded in your 1Ds2 for a 5D?
-
Hello Jack,
Thanks for the information, it seems interesting indeed.
What do you mean by "quality of the raw stitch"? This stitch is actually very easy in terms of images mapping since there are plenty of features. A bright wall is actually probably the easiest subject you could think of.
In terms of flat stitch and straight lines preservation, I have gotten similar results with PTgui, but I was working with a lens positioned on top of its nodal point.
What I find potentially interesting is the fact that camera was not positionned accurately on top of its nodal point.
However, to understand better the quality of the performance displayed by CS3 vs other tools, it would help to:
1. The focal lenght used,
2. Know how far you were from the wall,
3. Know how many mm away from the actual nodal point of that lens the camera was located.
Thanks for sharing these information if you have them.
Regards,
Bernard
-
Bernard:
1) 45mm (TSE)
2) Approximately 3 meters at the closest point, 15 or so from the farthest (lens was focused at about 4 or 5 meters)
3) I really have no idea -- but I suspect at least 20 or 30 mm "behind" the 45 TSE's nodal point.
What I meant by quality of the raw stitch is there is no artifacting on the fine detail as appears in the far grass in these jpegs -- if I crop and print the perspective correct version at 20x40 inches it appears as would a high-resolution single capture.
-
Ray:
I've only started using it myself and have done maybe half a dozen images with significant sky. In these, I have not been able to see a seam in the sky; blended skies look perfect right out of the gate. I actually play the "find the seam" game with most of these stitches, only to discover I was wrong when I toggle the layer off -- it is that good
Cheers,
-
I actually play the "find the seam" game with most of these stitches, only to discover I was wrong when I toggle the layer off -- it is that good
[a href=\"index.php?act=findpost&pid=101464\"][{POST_SNAPBACK}][/a]
Jack,
Looks like I'll have to find a way to download CS3 on a 56k dial-up connection. I've got quite a few images taken over the years for stitching purposes, which didn't work out to my satisfaction due to parallax errors. I even bought a big, heavy, Manfrotto pano head some years ago, big enoughto hold my RB67, with lots of complicated adjustments in relation to focal length and distance to subject, which I've hardly used because it's just too cumbersome and top-heavy. It seemed a good idea at the time, though .
Another area in CS2 which I think is lacking and which I hope has been improved in CS3, is merge to HDR. This is another situation where a tripod seems necessary. Although there is an 'automatic alignment' feature in 'merge to HDR', I have found it to be quite useless for hand-held shots.
-
Start the download and then go to bed.
I have to say that in the few tests I've run CS3 achieved panos that I wasn't able to before. Using full auto mode.
-
Do you still have to change the file from 16 bit to 8 bit to accomplish this in CS3?
-
Do you still have to change the file from 16 bit to 8 bit to accomplish this in CS3?
[a href=\"index.php?act=findpost&pid=101697\"][{POST_SNAPBACK}][/a]
Nope
-
Start the download and then go to bed.
[a href=\"index.php?act=findpost&pid=101611\"][{POST_SNAPBACK}][/a]
I've been downloading CS3 for the past 16 hours. Have about 7 hours more to go.
I'm using 'Free Download Manager'. I'll be interested to see if this 338MB file has become corrupted during the download.
Use of the computer for other purposes, such as making this post, becomes much slower than normal, of course.
-
Bernard:
1) 45mm (TSE)
2) Approximately 3 meters at the closest point, 15 or so from the farthest (lens was focused at about 4 or 5 meters)
3) I really have no idea -- but I suspect at least 20 or 30 mm "behind" the 45 TSE's nodal point.
What I meant by quality of the raw stitch is there is no artifacting on the fine detail as appears in the far grass in these jpegs -- if I crop and print the perspective correct version at 20x40 inches it appears as would a high-resolution single capture.
[a href=\"index.php?act=findpost&pid=101463\"][{POST_SNAPBACK}][/a]
Hello Jack,
Thanks for the quick reply.
We might have something interesting then. If I find the time, I'll try PTgui on your images to check how well it performs.
- I expect it to be able to be in the same league for planar projection (I have good results with a 35 mm lens on my D2x),
- I am not sure how well it will correct the nodal point error.
Regards,
Bernard
-
Jack, the single row pano looks very good. Will CS3 handle multi-row panos? Has anyone tried this?
Thanks,
George Deliz
-
Jack, the single row pano looks very good. Will CS3 handle multi-row panos? Has anyone tried this?
Thanks,
George Deliz
[a href=\"index.php?act=findpost&pid=101931\"][{POST_SNAPBACK}][/a]
Yes, and does it just as well as the single row above, but it takes it a little longer to process it
I should add a word on processing. While I find CS3 significantly faster than all other dedicated solutions I've tried, it still does take some time to process -- there is obviously a *lot* of math processing going on to make this work...
A friend of mine threw 12 Aptus 75 frames (in 16-bit about 240 megs each) at it (6x2 grid) and it took 4 hours to do the stitch on his power Mac -- but he said the result was perfect
I have thrown 8 5D 16-bit frames at it (about 70 megs each) and it took about 5 minutes on my Dual-Xeon (4-G ram) PC. Surprisingly, my duo-core centrino laptop (2-G ram) processed those same 8 frames in 8-bit in about the same amount of time.
-
A friend of mine threw 12 Aptus 75 frames (in 16-bit about 240 megs each) at it (6x2 grid) and it took 4 hours to do the stitch on his power Mac -- but he said the result was perfect
[a href=\"index.php?act=findpost&pid=101946\"][{POST_SNAPBACK}][/a]
Hi Jack F,
I have seen some of your images and it was clear to me that you certainly pay attention to detail. Do you personally have any Apturs75 experience, did you happen to see any CF issues with regards to your friend’s a75 images?
and it took 4 hours to do the stitch on his power Mac -- but he said the result was perfect
[a href=\"index.php?act=findpost&pid=101946\"][{POST_SNAPBACK}][/a]
I can’t believe it takes 4 hours to do the MF stitch processing. I was hoping that it could be done during the brake, more like 15min./30min./perhaps even 60min. but 4 hours? forget it, he is better of with Seitz 6x17 full frame with Linhof/Schneider Lenses. Basically, if I purchase P45 or H3D 39 it’s even worse so I better think about adding another mac dedicated for rendering I guess like my brother has in his Animation Department.
Oh, I hope your friend is wrong on this one.
Regards
Danijela
-
This speed problem, exists always with panos. So I decided last year, that I don't have the time to wait 2 hours to get the Panos rendered. I needed to do some other stuff in that time. Now I have a second PC connected with Ethernet 1000MB and it works well.
Ok I'm talking here about 20-40 1DsMk2 files, which are getting kind of big ;-)
-
This speed problem, exists always with panos. So I decided last year, that I don't have the time to wait 2 hours to get the Panos rendered. I needed to do some other stuff in that time. Now I have a second PC connected with Ethernet 1000MB and it works well.
[a href=\"index.php?act=findpost&pid=101988\"][{POST_SNAPBACK}][/a]
That is excactly what I was thinking, only you know Mac's are monsters but they do cost a a bit more and that is what bothers me at the moment. However, not many options, so you pay for what it is I guess.
Ok I'm talking here about 20-40 1DsMk2 files, which are getting kind of big ;-)
[a href=\"index.php?act=findpost&pid=101988\"][{POST_SNAPBACK}][/a]
Of course 1DsMk2 files are big enough for the second computer if you are processing 24 to 40 files.
I guess it will all work out fine, since batch processing on CaptureOne is fastest so far, but not fast enough. However, compared to the opponents it is substantialy fast, we'll see.
Regards
Danijela
-
Hi Jack F,
I have seen some of your images and it was clear to me that you certainly pay attention to detail. Do you personally have any Apturs75 experience, did you happen to see any CF issues with regards to your friend’s a75 images?
I do not have enough MF DB experience to express a significant conclusion on any of them, but will offer my initial impressions. In my limited comparisons of all the current high-end digital MF backs, I give the A75 the nod over the P45 and Hassy H3D-39 -- to my eye, the file is smoother and less "digital" in appearance. Howeverbut, all three of these backs are extraordinary performers and I don't think you can go wrong with any of them if they suit your purpose.
The A75 does not appear to show significant color-shift even when used with shifted glass on a view camera. By comaprison, the P45 clearly does show color shift. I have not yet seen files from the H39 back mounted on a view camera so cannot comment either way on it.
-
I've been doing a bunch of stitching lately and Im finding that PTGUI is pretty damn great.
I'm thrilled to be able to get salvagable grab shots when I don't have a tripod in my trunk. I'm wondering how much the PS3 stitcher lets you control? I'm still learning PTGUI but I got up to a decent speed pretty fast on it.
This shot is 4 frames stitched in a rectilinear projection. I was pointing way up and over etc..all hand held (that billboard is very tall and my camera position is about a street width away)). Using the vertical line control points (t-1 control points) is great for getting verts, vertical and messing with the horizon line and other stuff...you can get amazingly tight stitches with this app.
(http://i122.photobucket.com/albums/o274/rentChicago/a-houseWebV2.jpg)
-
I've been doing a bunch of stitching lately and Im finding that PTGUI is pretty damn great.
[/img]
[a href=\"index.php?act=findpost&pid=102282\"][{POST_SNAPBACK}][/a]
Yep, that is my feeling as well, but there are always some things to improve on. Evaluating PS CS3 stitching capability might be worth it.
Personnally, I think that it is a waste of resources for Adobe to develop such functions though.
Instead of trying to compete with - already excellent - small guys in niche applications, they should focus on their most important job which is to keep providing a top notch platform for image editing. We still don't have 64 bits support.
Regards,
Bernard
-
Yep, that is my feeling as well, but there are always some things to improve on. Evaluating PS CS3 stitching capability might be worth it.
Personnally, I think that it is a waste of resources for Adobe to develop such functions though.
Instead of trying to compete with - already excellent - small guys in niche applications, they should focus on their most important job which is to keep providing a top notch platform for image editing. We still don't have 64 bits support.
Regards,
Bernard
[a href=\"index.php?act=findpost&pid=102296\"][{POST_SNAPBACK}][/a]
Sorry but for what would you nned 64 bits ????
-
Sorry but for what would you nned 64 bits ????
[a href=\"index.php?act=findpost&pid=102338\"][{POST_SNAPBACK}][/a]
This really isn't the topic of this thread, sorry about that Jack. I'll answer once on this... I would like to have 64 bits support to save time when dealing with really large images.
- 4x5 scans with multiple layer masks,
- panoramas from Mamiya ZD...
I often end up having image size on file that is larger than 1.5 GB, sometimes in the 2GB range.
A recent post from Adobe PS's product manager acknowledged the value of 64 bits for images larger than 1GB. I will give a try to 4GB systems in my next box (either Mac Pro 8 cores when it is released or a Vista box), but I have seen no evidence that it will help enough with my size of images.
For your information, many professional applications, like CATIA V5 in the CAD area for instance, have been migrated to a full 64 bits implementation about one year ago already.
Cheers,
Bernard
-
This really isn't the topic of this thread, sorry about that Jack. I'll answer once on this... I would like to have 64 bits support to save time when dealing with really large images.
- 4x5 scans with multiple layer masks,
- panoramas from Mamiya ZD...
I often end up having image size on file that is larger than 1.5 GB, sometimes in the 2GB range.
A recent post from Adobe PS's product manager acknowledged the value of 64 bits for images larger than 1GB. I will give a try to 4GB systems in my next box (either Mac Pro 8 cores when it is released or a Vista box), but I have seen no evidence that it will help enough with my size of images.
For your information, many professional applications, like CATIA V5 in the CAD area for instance, have been migrated to a full 64 bits implementation about one year ago already.
Cheers,
Bernard
[a href=\"index.php?act=findpost&pid=102464\"][{POST_SNAPBACK}][/a]
Oh sorry my fault :-P Now I know what You mean and you are right. My question was than I just read it really quick, that you ment a 64 bit Image ;-)
-
Jack,
In your rectilnear output, the 'verticals' on the right side of the stitched image aren''t close to vertical...
-
It's not vertical in that last single frame either
Fact is, I wasn't all that careful setting up this image. Also, this was the side of a factory built out of brick in 1875. Could be the foundation has settled a bit or could be I had the camera at an angle.
-
I'm going to continue this thread rather than start another, to preserve continuity which is important. Jack has tantalising raised the issue that PS3's Photomerge is as good as any dedicated stitching program. Some reader have suggested the new CS3 Photomerge cannot compete with PTGui. I think this is probably true.
My favourite stitching program is Panavue's Image Assembler, which has continually improved over the years. I was therefore very interested in how PCS3's Photomerge would compete.
It simply cannot. I'm led to the conclusion there is no substitute for a dedicated, professional stitching program.
Okay! let's not denigrate the improvement of Photomerge in PCS3. It is improved (over CS2) and it is possible with just a few images, of the right sort, to produce a perfect stitch, automatically.
However, with big projects it doesn't pass muster. For 2 or 3 or 4 images though, it's sometimes quite good in automatic mode.
Most stitching programs have an automatic mode. Sometimes it works perfectly. More often than not, it doesn't.
How do we handle a series of images, in CS3 photomerge, that doesn't work in automatic mode? Not easily, I suspect.
It so happens I have a huge stitching program on my hard drive (and of course backed up on DVD) of a view of Brisbane from Mt Coot-tha, right out to the Bay on a clear day. (That's Brisbane, Australia. Not Brisbane, America ).
The shots were taken with a 20D and 400mm lens (100-400 IS) and the completed stitch will consist of 2 rows of 33 images each, ie a total of 66 images.
I tried stitching the first 16 images, top left segment, using CS3. It couldn't handle it. Zero result.
Below is the result from Image Assembler.
[attachment=1984:attachment]
This is not satisfactory, but at least the joins are perfect.
As I mentione PS3 could not handle 16 images, but it could handle 12. Here's the result.
[attachment=1986:attachment]
The joins are also perfect. But both of these images are no use to me. How could I print such images on a rectangular piece of paper. I'd have to crop them so severley, I'd have nothing much left.
There's a solution to this problem in Image Assembler. I select just 3 images from the row and create a 'lens type' using the 'Lens Wizard'. Very straightforward and quick. I then use this lens type for the whole 33 image stitch, but not unfortunately in automatic mode. I have to use 3 pairs of flags at each overlap. Positioning these flags is tedious to say the least, but it produces the goods, as can be seen below in the full top row of 33 images.
[attachment=1987:attachment]
Now, you should notice in the first 2 images, the sea is sloping dangerously. This could be catastrophic for the city of Brisbane.
In the third stitch, everything is just right. The sea is perfectly horizontal and minimal cropping is required for a rectangular print.
Accept my word for it; all joins in all 3 images are perfect.
But, there may also be another advantage of a professional stitching program. I think all of us who have tried these procedures are aware of the problems of subject movement during the time it takes to move the camera from one frame to the next.
Here's an example of such movement in the above stitch, which is a screen grab showing the positioning of the flags before running the stitch.
[attachment=1988:attachment]
As you can see, between one frame and the next, the yacht has moved and someone in a speed boat has appeared on the scene.
How will Panavue handle this? Will it produce a double image of the yacht? Not on your Nelly. It'll include all relevant image detail and exclude all duplication, as can be seen in a small crop from the final stitch below.
[attachment=1989:attachment]
Well, what more can I say. I think Panavue should send me a cheque for all this free advertising .
-
For the technically minded, I should add that the full 33 image stitch, in 8 bit, is 720MB and 68,884 pixels long. After stitching the 2nd row and perhaps replacing and extending the sky (which is a bit uninspiring due to pollution), I'd expect the completed image to be at least 1.5GB. At 240ppi, uninterpolated, it should print out about 24ft x 3ft.
The camera was the Canon 20D with 100-400 IS lens at 400mm, f16 and ISO 400.
However, my printer is only 24" wide so I'll have to downsample the image to 16ft x 2ft.
Here's a crop of the centre 2 to 3 images from the top row.
[attachment=1990:attachment]
-
Jack has tantalising raised the issue that PS3's Photomerge is as good as any dedicated stitching program. [{POST_SNAPBACK}][/a] (http://index.php?act=findpost&pid=104515\")
FTR, I don't think I actually said that here; what I said was I found it faster... If I was not being clear, my apologies. What I tried to imply was that the new CS3 capability in this arena is impressive, easy to implement while already working inside CS and thus certainly convenient -- and at the end of the day, more than adequate for my needs. I suspect it will be adequate for many other users too, though power-stitchers like you are excepted
Regarding your program's results, they are impressive Ray -- and obviously you are more experienced in assembling large panoramics than I am.
~~~
To be clear though, my reason for stitching is not generally to create a pano, but rather to regularly assemble 2 or 3 captures -- and on rare occasion 4, 6 or 8 -- for a higher-resolution normal aspect ratio single image. Sometimes I will certainly grab a 6 or 8 frame set to assemble for a pano, but that is not my normal cup of tea.
FWIW, I did experiment with really large panos when I had my Betterlight scanning back and dedicated Betterlight pano adapter. That device made panos up to 9000 pixels high by 360 degrees -- with a long lens this was enough image data to crush any modern computer's ability to process it. If I were interested in routinely making large panoramic images, this is the device I would use over attempting to assemble 32 frames after the fact. However, in the end, I only ever made a few "skyline" style panos I really liked. I linked to one below, "San Francisco at sunset", that prints uninterpolated to 17"x34" at 360PPI.
I also made some big ones -- 2-gig file size in 16-bit that would print uninterpolated to 2 feet x 10 feet at 360 PPI and obviously bigger (3' x 15') at 240PPI. What I soon learned is you need a pretty large wall to display them and not many folks have that kind of wall space... I also very quickly learned mounting them was a pain and I did not even have enough space to store the rolled prints!
Here is the SF skyline sunset pano link for those interested. In the 17" print, you can see the individual stripes in the American flags on the building tops, though it's tough to see the flags themselves in this 1/36th scale 1000x2000 pixel image:
[a href=\"http://jack.cameraphile.org/gallery/view_photo.php?full=1&set_albumName=album10&id=sunsetskyline2]SF Skyline Pano[/url]
Cheers,
-
Question any differnece between Panavue's Image Assembler ? Pano Tool Assampler ? or PTGui ?
-
I'm going to continue this thread rather than start another, to preserve continuity which is important. Jack has tantalising raised the issue that PS3's Photomerge is as good as any dedicated stitching program. Some reader have suggested the new CS3 Photomerge cannot compete with PTGui. I think this is probably true.
...
I couldn't be arsed to tinker with pano programs the way you do, especially since fully automatic stitching software exists.
I have a ~75-image 330 megapixel panorama of NYC nighttime skyline shot from Brooklyn Heights I'm working on. I used Autopano and all I did was drag-and-drop all the images into the program and a few hours later I had an almost perfectly stitched shot. There was only one problem with a corrugated sheet metal door near the edge of the shot I need to fix.
The point is that you can get all the functionality without having to do any control points, aligning or even arranging shots in their proper places. Drag, drop, go to bed. The shot's finished in the morning and you can start working on the actual post.
-
FWIW
Has anyone done or seen a really stunning image executed with a large mosaic stitch? It is a ton of work and I personally have yet to see an image done that way that was worth all the effort except as a demo or necessary for some documentation project. I do find some of Chris Jordon's work conceptually interesting "Running the Numbers" Chris Jordon (http://www.chrisjordan.com/).
-
FWIW
Has anyone done or seen a really stunning image executed with a large mosaic stitch? It is a ton of work and I personally have yet to see an image done that way that was worth all the effort except as a demo or necessary for some documentation project. I do find some of Chris Jordon's work conceptually interesting. See Running the Numbers Chris Jordon (http://www.chrisjordan.com/).
[a href=\"index.php?act=findpost&pid=104667\"][{POST_SNAPBACK}][/a]
Most problem of today is still post processing. I have some larger stichted images. around three rows of 10-15 1DsMk2 images, but it is such a pain to work with them I feel like ages ago. 300 or more MP are fantastic in detail and for huge prints, but with the current technology they are also a pain.
-
FWIW
Has anyone done or seen a really stunning image executed with a large mosaic stitch? It is a ton of work and I personally have yet to see an image done that way that was worth all the effort except as a demo or necessary for some documentation project.
That's a good point Kirk.
Just because it can be done does not mean it automatically creates a good image. Artistic basics still apply and vision is 90% of the final image IMO. The typically short duration of the sweetest shooting light or the rapidly changing high drama of a clearing storm do not lend themselves well to a mosaic stitch as too much changes between the first and last exposure. One might be able to make a few frames work, but it is unlikely one could make dozens of them work. Kind of leaves that use relegated to static images... I guess it's why some of us still like to capture images on large pieces of film
Cheers,
-
FWIW
Has anyone done or seen a really stunning image executed with a large mosaic stitch? It is a ton of work and I personally have yet to see an image done that way that was worth all the effort except as a demo or necessary for some documentation project. I do find some of Chris Jordon's work conceptually interesting "Running the Numbers" Chris Jordon (http://www.chrisjordan.com/).
[a href=\"index.php?act=findpost&pid=104667\"][{POST_SNAPBACK}][/a]
Excellent point, Kirk. Reminds me of a comment Minor White made back in the 1960s (when "archival processing" was just becoming all the rage): "What we need is more archival images, not more archival prints."
Eric
-
Eric,
Good point. I tell me students not to go overboard with things like archival processing of prints (film ok) until they have some mature work that deserves it. Otherwise it is a big waste of time and resources.
-
Eric,
Good point. I tell me students not to go overboard with things like archival processing of prints (film ok) until they have some mature work that deserves it. Otherwise it is a big waste of time and resources.
[a href=\"index.php?act=findpost&pid=104691\"][{POST_SNAPBACK}][/a]
Actually, for many years I processed all of my prints archivally, because once I had the system down, it was easier to stick with one "workflow" than to keep switching back and forth. But finding the archival images has always been harder.
I do plan to try stitching soon, but I can't see myself ever wanting to put more than about three images together. The amount of time and effort put into it should be appropriate to the potential quality of the resulting image, IMHO.
-
Hello Kirk,
I had the privilege to view, if I recall correctly, a 15 image multiple row stitch by Joseph Holmes at the most recent Digital Fine Art Summit by Alain and Natalie Briot. I believe the data was acquired with a Canon 5D and the 70-200mm lens at 100mm, I could have some of the details incorrect. What is important is that the image was stunning in composition and detail. The light was open shade and the scene was of a mixture of hardwoods in a hillside forest. The print was certainly worthy of "fine art" designation in my books and a testament to what can be accomplished with today's equipment. He used the panorama gear from Really Right Stuff which motivated me to purchase the same gear and begin my own investigation of the technique. I don't know if that print is on his website yet, but it was stunning to see.
Jeff Ball
FWIW
Has anyone done or seen a really stunning image executed with a large mosaic stitch? It is a ton of work and I personally have yet to see an image done that way that was worth all the effort except as a demo or necessary for some documentation project. I do find some of Chris Jordon's work conceptually interesting "Running the Numbers" Chris Jordon (http://www.chrisjordan.com/).
[a href=\"index.php?act=findpost&pid=104667\"][{POST_SNAPBACK}][/a]
-
To be clear though, my reason for stitching is not generally to create a pano, but rather to regularly assemble 2 or 3 captures -- and on rare occasion 4, 6 or 8 -- for a higher-resolution normal aspect ratio single image. Sometimes I will certainly grab a 6 or 8 frame set to assemble for a pano, but that is not my normal cup of tea.
[a href=\"index.php?act=findpost&pid=104578\"][{POST_SNAPBACK}][/a]
Jack,
I've done a few more tests using CS3 Photomerge to stitch just a few images at a time and it doeas appear to outperform Pananvue's Image Assembler in auto mode in these circumstances.
The following 3 handheld shots with the Sigma 15-30mm and 5D are impossible to stitch in Panavue's auto mode, but CS3 does just as good a job as the more tedious 'flag positioning' method that I have to use with image Assembler.
[attachment=1996:attachment] [attachment=1997:attachment] [attachment=1998:attachment] [attachment=1999:attachment]
However, the mountain range on the far right is sloping down a bit too obviously, so I tried using the 'perspective' option in Photomerge and it did a wonderfully abstract job consisting of a triangle and a rectangle which I quite like. It sort of almost conforms to the rule of thirds, but not quite, which makes it more interesting .
[attachment=2000:attachment]
-
The point is that you can get all the functionality without having to do any control points, aligning or even arranging shots in their proper places. Drag, drop, go to bed. The shot's finished in the morning and you can start working on the actual post.
[a href=\"index.php?act=findpost&pid=104589\"][{POST_SNAPBACK}][/a]
I'm always willing to try a faster method. The problem is, there's usually a learning curve with new software, but a program that claims to handle things automatically shouldn't take too long to get the hang of.
I downloaded the trial version of Autopano and tried it out on the 33 images that Panavue handled flawlessly with flags. It seems to do a pretty impressive job automatically aligning the images in a straight line, but I can't find a way to enlarge the images to full detail so I can't inspect the joins. The trial version just gives me a thumbnail preview of the final stitch and the 'save' function is disabled. Perhaps there's some obvious control that I've missed. I can't understand why anyone would offer a free trial without an option of seeing the full rez stitch.
Another gripe is the way it handles color profiles. All my images have an embedded ProPhoto RGB profile. The stitched images in Autopano all appear in sRGB.
Although the images seem well aligned in a straight line, the color is seriously off. Some of the stitches were really bad. Below is the best of them, compared with the panavue stitch which I've copied and pasted below. As you can see, the tone and color changes from left to right to a degree that is not acceptable.
I tried various setting, using maximum quality and maximum number of detection points etc. I don't see what else I can do, unless you've got any suggestions?
[attachment=2001:attachment]
-
Ray:
FWIW, here is my version of your three frames above using the "interactive" option, then checking "perspective" after I confirmed positioning for CS3's Photomerge. Maybe not as good as your dedicated program, but as I said above, still pretty darn good IMO -- and it did this in about 10 seconds with those jpegs
All I need to do is level the horizon in CS, and maybe transform perspective a bit to suit, and then I think it woud be quite good for a casually captured scenic pano:
-
Okay, I spent a few more minutes on Ray's pano. Here it is after a quick horizon level, slight perspective transform, crop and minor curves adjustment in CS3:
-
Here's another vote for CS3's photomerge. I recently tried it and I was very pleased with the results, especially considering that the process was so easy. The program definitely has its limitations in that a stitch I tried directly from raw had blown skies while the same stitch from pre-processed .psd's was perfect. A difficult series that I took a couple of years ago with the camera in AE, AF mode simply didn't stitch properly. The same series after some manual blending did stitch perfectly, something I have yet to achieve from other stitching programs. For my purposes, a few shots to produce higher resolution for printing up to 16" X 24" or perhaps an ocasional 24" X 36" for kicks, it looks like CS3 will be all I need, especially if the final version is improved over the beta.
George Deliz
-
Okay, I spent a few more minutes on Ray's pano. Here it is after a quick horizon level, slight perspective transform, crop and minor curves adjustment in CS3:
[a href=\"index.php?act=findpost&pid=104802\"][{POST_SNAPBACK}][/a]
Jack,
Thanks for the attempt. Have you noticed you lost a stretch of mountain range in the left segment . Also the mountains look smaller in general, in relation to the rest of the image. I took the same image from Photomerge's auto stitch and used 'free transform/warp' to pull up the two top corners, as below. You also lost the people. Were those cloned out? I'm a bit undecided about this. People have a tendency to walk up close to a large print on the wall print to see the detail. When they do with this one, they'll notice a group of tourists taking photos of each other with their P&S cameras. Leaving them in might be a sort of social commentary .
[attachment=2006:attachment]
-
Jack,
Thanks for the attempt. Have you noticed you lost a stretch of mountain range in the left segment . Also the mountains look smaller in general, in relation to the rest of the image. I took the same image from Photomerge's auto stitch and used 'free transform/warp' to pull up the two top corners, as below. You also lost the people. Were those cloned out? I'm a bit undecided about this. People have a tendency to walk up close to a large print on the wall print to see the detail. When they do with this one, they'll notice a group of tourists taking photos of each other with their P&S cameras. Leaving them in might be a sort of social commentary .
[attachment=2006:attachment]
[a href=\"index.php?act=findpost&pid=104816\"][{POST_SNAPBACK}][/a]
Ray, I simply cropped to the maximum rectangle after orienting the horizon and AFTER generating the PERSPECTIVE CORRECT rendering per the first version I posted.
If you look at that image, you can see that the perspective correction squeezed out the people and the rectanglar crop of that image would eliminate part of the horizons and sky at the edges...
No way around that unless you want to stick with a cylindrical projection or reposition only projection. These both will have the resulting curved horizon, since it appears your camera was angled down slightly as you panned.
Here is an uncropped, unprocessed version of a "reposition only" CS3 assembly. You may prefer it since the people are there and the mountains are there, just the horizon has that swing-lens curve which I don't care for. Different strokes :
-
Lastly, here is the "Cylindrical" or pan projection from CS3, uncropped and unprocessed, that attempts to correct for a level pan, so it pulled the lower corners in a bit and bowed the captures for a more correct cylindrical projection.
Again, my point is simply that CS3 can do a credible job with relative ease, even on difficult stitches, and yet offers a variety of options for assembling them at the same time:
-
FWIW
Has anyone done or seen a really stunning image executed with a large mosaic stitch? It is a ton of work and I personally have yet to see an image done that way that was worth all the effort except as a demo or necessary for some documentation project. I do find some of Chris Jordon's work conceptually interesting "Running the Numbers" Chris Jordon (http://www.chrisjordan.com/).
[a href=\"index.php?act=findpost&pid=104667\"][{POST_SNAPBACK}][/a]
Actually Jack's image of SF seems to qualify.
Edmund
-
Well, the one I find most interesting of all of these so far is Ray's "Triangle plus Rectangle!" But perhaps it belongs in the "But is it art?" thread.
-
I'm pretty sure Chris Jordan shoots most of his work with an 8x10 view camera..not to sure about the "running the numbers" series but I do believe he's an 8x10 gilm guy. I don't think he's stitching stuff.
-
Actually Jack's image of SF seems to qualify.
[a href=\"index.php?act=findpost&pid=104870\"][{POST_SNAPBACK}][/a]
A nice shot of SF with a lovely red sunset glow. If the details on the flags are visible, it must be really high resolution. But I'm not sure it qualifies for a 'large mosaic stitch'.
As Jack mentioned, this prints to 17"x34" uninterpolated at 360ppi. It's not clear what camera Jack used, but a vertical shot with the 1Ds2 should produce a 14" high print at 360ppi. I think Jack uses higher rez cameras than the 1Ds2.
-
Pete, Chris is shooting 8x10 and stitching with some images in that series and before. He has recently sold off most if not all of his LF stuff and is switching to digital.
-
Lastly, here is the "Cylindrical" or pan projection from CS3, uncropped and unprocessed, that attempts to correct for a level pan, so it pulled the lower corners in a bit and bowed the captures for a more correct cylindrical projection.
[a href=\"index.php?act=findpost&pid=104827\"][{POST_SNAPBACK}][/a]
Jack,
I've probably botched the image preparation for this stitch (just as I botched the taking of the shots ).
I usually try to make the job for stitching programs easier by cropping all images to the same size and rotating the crops (if necessary) to bring elements in each frame to a similar level. With these handheld wide-angle shots, the levels were all over the place. How did I manage that? Well, I certainly wasn't drunk. Perhaps I was just exhausted after climbing an 800 metre hill before breakfast.
I tried the stitch again with the 3rd image uncropped and larger in size. It worked just fine and gave me more image on the right. This is another 'tick' for photomerge.
[attachment=2030:attachment]
I'm particularly pleased that the large, red lens-flare spot on the left of the third image (to the right) has disappeared in the stitching process.
[attachment=2031:attachment] [attachment=2032:attachment]
This is a known common problem of this lens. Quite often I just raise my hand to protect the lens. A rectangular piece of cardboard would be better, though.
It seemed a pity to crop out that extra image area bottom right, so I searched for other similar images of the same scene and did a copy/paste and clone job in the bottom left corner.
The near-final rendition, which is almost ready to print, appears below.
[attachment=2033:attachment]
-
Edmund, thanks for the compliment on my SF skyline shot! But as Ray pointed out, it does not qualify as it is not a stitch but rather a single scan capture from a Betterlight back.
Ray, that last un-cropped full frame is clearly superior. I have never bothered to try and "help" the program -- it is way smarter than I am
Cheers,
-
Well, the one I find most interesting of all of these so far is Ray's "Triangle plus Rectangle!" But perhaps it belongs in the "But is it art?" thread.
[a href=\"index.php?act=findpost&pid=104902\"][{POST_SNAPBACK}][/a]
Eric,
We seem to think alike . I also thought this was quite cute. I'm seriously considering printing it, if for no other reason than to study the non-plussed expressions of those who view it .
-
Pete, Chris is shooting 8x10 and stitching with some images in that series and before. He has recently sold off most if not all of his LF stuff and is switching to digital.
[a href=\"index.php?act=findpost&pid=104935\"][{POST_SNAPBACK}][/a]
Another one bites the dust. I wonder how it'll change his approach to things. Definitely a look to his work and prints that was heavily wrapped up in a view camera mind set. I liked that.
-
I have never bothered to try and "help" the program -- it is way smarter than I am
[a href=\"index.php?act=findpost&pid=104940\"][{POST_SNAPBACK}][/a]
That's for sure, but I don't believe in miracles .
When I know the images are 'bodgy', I feel compelled to help the program.
Here are the 3 uncropped images at the basis of the above stitch.
[attachment=2034:attachment] [attachment=2035:attachment] [attachment=2036:attachment]
Below is the automatic stitch of those 3 images.
[attachment=2037:attachment]
Now, I know what you are thinking. Was this guy on prohibited drugs when he took these shots? No, I wasn't. I wasn't on any drugs of any description. I didn't even have a cup of coffee before I set off on the pre-dawn climb.
The automatic stitch of these 3 images is impressive. I cannot fault any join. But the sloping horizon is a signifiant problem. That's why I sometimes try to help the program, Jack .
-
Well I suppose it depends on the original captures and guess it makes sense in that the GIGO laws still apply But it is good to know that there may be salvation for a poorly-captured set of stitch images -- thanks for sharing the full details!
PS: However, the second set of frames look like a completely different set than the first... That first frame with the people in it has much less foreground in it and the second frame has the people which the original set did not. Edges are different too. Maybe you grabbed a couple of sets and got them mixed up?
Cheers,
-
I didn't even have a cup of coffee before I set off on the pre-dawn climb.
That's the problem, Ray! I could certainly never get a pre-dawn horizon straight if I hadn't had my morning coffee. In fact, I'd probably leave the lens cap on and not even notice it.
-
Good point. I tell me students not to go overboard with things like archival processing of prints (film ok) until they have some mature work that deserves it. Otherwise it is a big waste of time and resources.
[a href=\"index.php?act=findpost&pid=104691\"][{POST_SNAPBACK}][/a]
Careful, as you don't always realise the value of things at the time.
People can become famous/infamous later and suddenly those arty B+W nude pics you did of Madonna Ciccione are worth a fortune.
-
-- thanks for sharing the full details!
[a href=\"index.php?act=findpost&pid=104996\"][{POST_SNAPBACK}][/a]
Jack,
You want the full details? Okay! I'm in strict scientific mode for this post; no arty farty pretensions.
There's no doubt that the above 3 images were taken in progressive sequence, from left to right. The metadata indicates there are just a few seconds between each shot. However, looking at the larger sequence of shots on each side of these 3 (I took a few dozen or so from this position with various lenses), it's now clear that the first image is the third of a previous group, panning from right to left, not the first of a second group panning from left to right. I took over 12,000 images on this trip and have difficulty sometimes in remembering the precise circumstances of each shot. However, having given it some thought, I can now reconstruct the exact circumstances.
These particular images were taken with the 5D on a Manfrotto 740SH tripod from the top of a 'look out' tower. It's a rather inadequate tripod for stitching purposes, being a ball-head, and doesn't have much height, even with the column fully extended. But it's very light and compact and serves its purpose for stitching with a T&S lens using the RRS L bracket and clamp.
The 'look out' tower of course has a fence around the platform to prevent people falling off. My tripod was not high enough to allow the lens an unobstructed view over the top of the fence. My guide, a very obliging chap, suggested he might be able to find a few slabs of rock to elevate the tripod. So whilst my guide was searching below for suitable slabs of rock, I was taking handheld shots, some for stitching and some not.
A few minutes later the guide appeared carrying a few rocks that did the trick. I normally pan from left to right because Panavue IA positions images in numerical sequence. But on this occasion, lens flare from the sun on the right was a problem, so I started from the right, then back from left to right, then back from right to left and so on. Here are the first 5 shots.
[attachment=2044:attachment] [attachment=2045:attachment] [attachment=2046:attachment] [attachment=2047:attachment] [attachment=2048:attachment]
As you can see, the first image is reasonably level, judging by the clouds. The second image is not too bad but has a more noticeable slope to the left and the 3rd image is just atrocious. To what extent this is due to the ball-head and to what extent due to the effects of a wide-angle lens tilted down, I'm not sure. But it's clear that between the third and fourth shot, I tried to correct the tilt but overcorrected so the horizon is now sloping the other way slightly. However, in the 5th shot, the clouds are sloping in the opposite direction to the 4th shot, which tends to indicate the ball-head is not keeping the camera level.
Now, as you know, a good tradesman always blames his tools when things go wrong, so I'm blaming that ball-head tripod.
I converted the first 3 images again, paying particular attention to keeping the vignetting amount, vignetting mid-point, temperature and tint the same for each shot, something I sometimes overlook. Keeping the conversion settings identical in every respect seems to improve the tonal transitions between images in the sky. The image below shows how CS3 auto stitch compares to the best I can do with Panavue's Image Assembler using flags and lens selection, and also Panavue in full auto mode.
[attachment=2049:attachment]
It's interesting to note that the CS3 result is significantly smaller in spite of more white background. The Panavue result in auto mode has duplicated the person being photographed in the foreground, otherwise it seems okay. The 'flag' stitch has a slight discontinuity near the base, as shown in the interpolated crops, and has used the duplicated figure in the middle image as opposed to CS3's preference for the first figure.
[attachment=2050:attachment]
I should also mention that the first 3 images in this sequence were taken with the zoom lens set at 21mm. When I corrected the horizon level in the 4th image, I also changed the focal length to 25mm. The stitches in previous posts consist of one image at 21mm and two at 25mm.
-
Makes sense Ray, I understand getting images from different sets confused. I often take several "sets" of one subject for a stitch if I think the composition warrants it. After all, once you're set up, taking the second or third set is no biggie.
The first time I did this, I was instantly "confused" when editing the images for the stitch -- was that 3 frames or 5 I took for this one, did I pan left to right or right to left and where does the best series start and end? Like you, I had to go back to the metadata, try and remember what I did when I shot it, generally play detective and work it all out. For those who have not done this, no it isn't as easy as it sounds if you look at just the order they show in the browser -- multiple sets in front of you can create confusion with many compositions, especially if you are processing them several days after you captured them.
So FWIW, I now use a simple trick. Since I'm in manual mode anyway, I just stick my hand in front of the lens and grab a frame between each set. Makes it easy to differentiate them after the fact...
Cheers,
-
So FWIW, I now use a simple trick. Since I'm in manual mode anyway, I just stick my hand in front of the lens and grab a frame between each set. Makes it easy to differentiate them after the fact...
[a href=\"index.php?act=findpost&pid=105243\"][{POST_SNAPBACK}][/a]
Jack,
That's a good trick. I'll try that next time. But ultimately, I'm going to stitch the images that produce the best picture, whether or not they are part of the same set, if it's possible, and it certainly seems to be possible... a 21mm shot with seriously sloping horizon and two 25mm shots with a different levelling of camera does not seem to be beyond the capabilities of CS3's Photomerge. If I can copy and paste a few patches from other similar images to fill in the corners, then I'll do it to get the result I want.
-
Just found this thread. Impressive results. Thanks for the detailed analysis. Considering the previous version - virtually unusable - this looks great. The skies appear perfect, quite often a problem in my experience. Have you found anything that breaks CS3's stitcher?
Peter
-
Jack,
That's a good trick. I'll try that next time. But ultimately, I'm going to stitch the images that produce the best picture, whether or not they are part of the same set, if it's possible, and it certainly seems to be possible... a 21mm shot with seriously sloping horizon and two 25mm shots with a different levelling of camera does not seem to be beyond the capabilities of CS3's Photomerge. If I can copy and paste a few patches from other similar images to fill in the corners, then I'll do it to get the result I want.
[a href=\"index.php?act=findpost&pid=105254\"][{POST_SNAPBACK}][/a]
I do the same except a hand with one finger at the beginning of the series and a hand with two fingers at the end. Helps to tell when one pano series starts and ends.
Very interested in trying multi focus stitching (any particular tricks to that one?)
-
I shoot my hand as an identifier, too. Years of film shooting has me accustomed to shooting "slates". These can be very good time-savers later since they isolate groups of frames.
As others have mentioned, horizontal panos are best shot left to right, vertical panos top to bottom. I will sometimes shoot a pano right to left if the rightmost frame is time-critical and the others aren't. I'll wait for a sailboat to cross the sunpath (or whatever) in the rightmost frame, then shoot the other frames once I have that one recorded.
I believe iView Pro will sort images that are shot close together in time. ie "give me a listing of all groups of images with less than five seconds between frames" Could be useful for culling pano source images from large shoots at a later date. I've yet to try this, though.
-
This speed problem, exists always with panos. So I decided last year, that I don't have the time to wait 2 hours to get the Panos rendered. I needed to do some other stuff in that time. Now I have a second PC connected with Ethernet 1000MB and it works well.
Ok I'm talking here about 20-40 1DsMk2 files, which are getting kind of big ;-)
[a href=\"index.php?act=findpost&pid=101988\"][{POST_SNAPBACK}][/a]
i dont have a 1DsII to test with, but this afternoon i just did an autostitch of 50 jpg images from a 20D, and it took about 35 minutes on a Mac Pro 2.66.
so i guess stitching 20-40 1DsII files probably takes 1.5 to 2 hours at most.