A detail of the test image showing that the size of buildings/objects such as the clock tower are very dependent on your location.
I also realised that blending the sky was going to be a complex issue if I wanted to print the image at a very large size.
This was the point where I confidently told the gallery that the exhibition would feature a print around fifty feet long and 44" high.
The sizes were limited by the gallery space (70 feet max. on one wall), the printer capabilities (44" maximum print width) and the printer maximum print length (18 metres, according to Canon specifications)
That gave me two months to get it to work...
Image capture and stitching
In capturing the images for the panorama segments, I needed to choose which lens to use.
A longer focal length produces more image detail, but would take longer to produce the shot. A wider lens would be quicker, but the resultant image might lack detail at the size I wanted.
Having decided that the print would be at least 300ppi resolution, I settled on the Canon EF50/1.4 lens - actually at around £300, my cheapest lens. However, at f/8 and using the geometry and vignetting correction of Adobe Camera Raw for processing my RAW files into 16 bit TIFF files, it performs extremely well.
With several days of good weather forecast, I decided to shoot at dusk - well known as a great time for architectural photography, where there is still natural light, but at a similar level to usual building lighting.
The camera used was my Canon 1Ds3 (21MP) at 100ISO, with 0.5 second exposures at f/8 for each shot.
After processing the RAW files at identical settings (small amounts of highlight recovery and shadow fill added) into TIFF files, the sequences were imported into Autopano Giga.
Note the slightly odd colour.
This is because I've converted the files into the large ProPhoto colour space and Autopano ignores this during its processing.
Why the large colour space? With the wide range of artificial lighting and strongly coloured sky, I want to keep the full range of colours in gamut. I know that my Canon iPF8300 can cover many of these deep red and blue colours with ease, and I'm also looking to minimise clipping of any channel.
I know that if I'd worked in Adobe98, not many areas of the image would show any problems, but the extra 'room' in working in ProPhoto at 16 bit was a benefit without any obvious downside - in this particular image.
The overall look of each sequence is very dependent on where the camera is located. Moving it only a few feet easily alters the relative heights of buildings.
Autopano Giga makes excellent use of all the (effective) processors in my Mac Pro, as shown by this activity monitor screen grab. All 16 cores are working pretty much flat out.
Eventually I had a series of stitched panoramas, which I could then place in order to make the full size panoramic sequence photo.
However, this is the point where automation stops.
Each segment needs to join along some naturally occurring seam.
Due to where the seams occur in the segment, they may not actually be the same height (in pixels ) in different segments.
This is where taking the images at higher resolution than strictly needed is of benefit.
Where there is a seam, I'd measure the height in pixels in each segment of an object at the seam.
If for example a wall is 6900 pixels high in one segment and only 6100 in the segment that it is to be matched to, then one or both values need changing.
Fortunately, with the cylindrical projection I've used, transforming the image (Photoshop transform>warp) is quite easy with the adjustment splines, you can see below.
Deciding which images to bend and which ones to leave, is very much a judgement to make, based on how the physical buildings are arranged and how you want to flow one panoramic segment into the next.
This really is the bit where your own skill and judgement is what counts. This is the bit that sets a large part of how the final print will feel as people walk along it.
The large amount of detail in the image is deliberate - it's there to draw the viewer in, to give the feeling that there is more to be seen, new secrets to be found.
Large images are usually pretty low on detail - they are meant to be viewed from a distance. This one is meant to be viewed at a distance -and- up close.
Eventually I ended up with six files (note the .psb file extension) and saved scaled (warped) versions of each.
At this stage I've carried out no colour or brightness correction - so far it's been about image geometry and composition.
I've next created a very big empty document, and then placed each layer onto it.
No need for any precise cutouts at this stage, since the warping process means that everything will match eventually.
With two layers in place, I'm already appreciating the fact that I'm using Photoshop CS5 in 64 bit mode and with 24GB of RAM in my computer.
This detail shows the placing of the second layer.
The edge of the second layer is cut away for the join. I need to plan somewhat carefully what I'm doing, since I've reduced the number of history states in Photoshop to make working with huge images somewhat easier. One accidental click can easily take long enough enough to recover from, to go and make a cup of tea...
I also made sure there was some 400GB of free space on my Photoshop scratch drive.
Regular saving of work is always important, but when working with lots of layers and a 23GB file, expect the save progress bar to hang around for enough time to make coffee and have a bit of a read of the paper.
The files are all in 'large document format' - 16 bit working and a file some 170k pixels wide are just going to slow things down.
People in the picture
One issue with any long exposure shot, and especially a composite image like this, is what to do about people wandering into the frame.
There is a pause function on the GigaPan, which will pause the capture process and then you can start again with a new shot at the same point.
If you look at the images above you can see more distant relatively stationary people and a few 'ghosts' of closer ones.
I've got 'clean' versions of many of these parts of the image, but despite my usual instinct to 'remove' inconvenient people from architectural shots, I decided that the ghosts gave a sense of movement to the scene that complemented the movement you get from the changing viewpoints.
Comments I've received from visitors to the exhibition have been favourable in this respect.
With longer exposures, there would be even fewer signs of people, or if you look at the daylight test versions of the picture earlier, you can see how lots of people changes the whole feel of the image.
- As an aside, I'd note that when I do panoramic seascapes, I invariably shoot hand held and quite quickly. Steadily advancing waves can make for quite appreciable stitching errors.
The first version of the assembled image wasn't going to be long enough, so I went out and shot some more potential panoramic segments.
Fortunately, with light activated street lights and clear skies, I was able to use the timing of the street lights to ensure equivalent lighting conditions over the three nights it took to get all the images I wanted.
Suffice to say, I now know how I'd plan a similar sequence photograph, if a client wanted one!
I'm now at a point where I have the image assembled, and a good feel for how it is going to look.
I'll not go into all the details of the adjustments that I made to the different sections, but the two main areas of concern were evenness of the sky and balancing the perceived brightness/contrast of each segment.
I'm very aware of the fact that everyone has their own favourite approaches to achieving results in Photoshop. I use a lot of masked layers, others may prefer a different technique.
The Photoshop layers display at the right gives an idea of the approach I took - this was part way into the work, the final version of the file had lots more...
The main layers have adjustment layers, some of which are masked.
I've balanced lighting at time of capture and in RAW processing, so most adjustments here are relatively minor.
On a few occasions I've duplicated small parts of a segment to add back parts that have been erased as part of another process.
Without covering click by click details it's difficult to show all this, and I freely admit that I learnt new Photoshop techniques along the way.
I worked a whole day on the first version of the image, only to go out that night, shoot some alternate segments and start again from scratch - the second version was simpler, quicker to produce, and looked a whole lot better.
Getting an even, but accurate sky was very important - it helps tie the whole image together.
If you've never done it before, take a sequence of images (camera on manual) just after sunset on a day where the sky is very clear.
Take the photos around the full horizon if possible, but try and ensure that you get the direction the sun has set, and the opposite side of the sky.
Now process the RAW files at identical settings and look carefully at how the colour varies from top of the picture to the horizon. Look too at the change in brightness, and if there are buildings around, how the different light from different directions illuminates them.
The background layer to the whole image is a series of graduated fills, based on the actual sky, looking down some of the streets in the image.
The example below, shows where I've cut away part of the sky at the Santander bank to show the graduated background layer, itself based on the 'real' sky to the left of the bank.
If I turn off the background, you can see that there is the problem of what to do with the lamp and it's flare rays.
The important thing to note is that the lamp and rays are brighter than the sky.
This means that if I cut out a chunk of image from the original layer, paste it in place and then brush in a mask with a lighten blending mode, then the flare rays will appear on the (background) sky.
In the final image, this small layer also has an attached curve to fine tune the effect.
There are lots of little bits around the picture like this. I'm sure if I did it again, it would be less complex, but that's one of the benefits of learning more about Photoshop during the process.
Remember that I didn't create the image just to produce a Photoshop tutorial ;-)
Here's the background layer, with its graduated fills, that's used to even up parts of the sky from the segments, particularly at the seams. The fill sections have all had 0.5% Gaussian noise added to reduce any potential for banding, and to more accurately represent clear sky areas of the original component images
Move your mouse over the image to see the actual picture.