June 17th, 2008, 07:14 PM
Stitching Panos is a challenge: a lot of work but a thrill when done well.
However, some issues:
It's easy to have more images than one needs and slow the process! Some methods are faster, but which works best for you. Sure one takes the images with M settings. Still sometimes skies wont match. So what do you use. What your order of blending and stitching. How many stops do you cover with how many layers?
Let's see your best.
Max 700-1000 pixels wide. Add 100% cut out if needed for details or issues.
June 18th, 2008, 02:03 PM
Let's see your best.
Not my best, hence the title of my post, yet perhaps something to learn from (I did).
Last Friday (the 13th!), I decided to drive home following a more scenic route than the one I had taken earlier to my destination. I set my on board navigation system to "Shortest Route" hoping/expecting it would guide me along the Southern dike that borders the Rhine river which almost cuts the Netherlands in half as it runs East to West to end in the "North Sea". As I was indeed being directed to the river banks, I suddenly saw a nice spot with amazing clouds rushing by due to a stiff wind (rain was forecasted later that afternoon).
Because one isn't allowed to park next to the road outside of a municipality, except for an emergency, I knew I had to be quick. I ran the risk of either getting a ticket for illegal parking, or getting hit by a sloppy driver on that narrow meandering road on top of the dike. So I hit the emergency blinkers and pulled over to a grassy strip beside the paved road, quickly got out of the
car, took my tripod with RRS pano contraption out of the back, placing it next to the car, leveled the ballhead, and mounted my camera with a TS-E 45mm lens on it. I grabbed my PalmPilot and looked up the offsets needed for that lens, which I had determined in advance, and adjusted the entrance pupil offsets.
I was torn between capturing the fascinating clouds and the interesting landscape, but decided that that might be interesting to convey in the image, 50% land, 50% sky. One has to remember that because of the general flattness of the Dutch landscape (a large part of it lies below sea-level, some reclaimed from the sea, and is slowly 'sinking'), one rarely gets a nice elevated vantage point, so even a look down from a dike could cause a mild sensation of vertigo with some
people ;-). Seeing as much interesting land as sky, instead of a narrow strip of land and a lot of sky is something special already. So, while looking over my shoulder to see whether I was still safe, I quickly decided to make a 2 row stitched HDR panorama of both grassland and sky, and sort things out back home.
I set the camera to manual exposure with 5 exposure bracketed shots (the 1Ds3 allows to select 2,3,5 or 7 shots), 1 stop exposure time difference between each, for each tile. Based on my experience (and no time for extensive spotmetering), I estimated that that would be enough to cover the full dynamic range of the scene, which was sweeping horizontally almost from back-lit to frontal-lit.
To capture the detail of the down-slope in the relatively flat terrain, I tilted the T/S lens forward to get the elevated foreground and the medium range below in the same focus plane, and stopped down to f/8 (introduces only a little diffraction blur) to gain a bit of vertical DOF towards the horizon. At that time a coach (bus) and another vehicle tried to pass each other in opposite directions exactly where my car was 'parked' (with 2 wheels slightly on the asphalt and 2 in the grass) in the bend of the meandering road on top of the dike. All I could do was hope for the best, and mentally prepare to take a dive with my tripod mounted rig.
I quickly set the camera to mirror lockup, and made the lower first row of bracketed exposures, while aiming for a small horizontal overlap between the portrait oriented tiles of the land area. I then swivelled the camera up for a row of cloud-cover shots with an overlap between both rows. I changed the base focus, tilt, and exposure time for the bracketed sky shots, hoping that the blending stage of the full pano stitch would also blend between the vertical exposure differences.
I took 85 exposures, which goes pretty fast when you only have to lower the mirror, reposition, and hit the exposure button (with 2 second selftimer delay) between each sequence of 5 exposures. I then quickly put the tripod in the back of the car, while taking the camera and my "Hyperdrive Colorspace" to the front. I placed the CF card in the Hyperdrive and put it on the seat beside me as it started backing-up the images. And back on the road I was, all in one piece, phew.
When I got home and saved the files to my Raid mirrored drives, upon closer inspection I started noticing the errors that had crept into my rushed workflow (almost makes one believe the Friday 13th superstition). The most disturbing to me was that the focus was not good enough for a full size pin-sharp rendering. There were some unwanted HDR ghosts in the edges of some clouds, despite Photomatix's removal attempts, and what's worse, I should have taken 2 additional rows of exposures for a better choice of the final crop.
The cumulation of needing more rows and the postprocessing needed to remove the ghosting artifacts, made me decide to try a quick stitch, and see what could be done with the material I had. So I chose only 17 of the best exposed tile positions, 8 for the land row and 9 for the sky row, and let "Autopano Pro" figure out how much could be done with it. The single exposure tiles surely would show no ghosting, the RRS pano gear would allow accurate stitching, and the different exposure times for land and sky would hopefully be blended by the Smartblend engine.
This is what resulted:
Because I was standing in a bend of the road, there were no straight line projection distortions to worry about. I could stitch this 162 degree horizontal FOV with a circular projection without any stretching of the corners of this wide panorama. The blending between the different single exposures allowed to span a larger dynamic range than would have been possible with identical exposure times. Blending indivdual tiles with varying exposures could be seen as a poor man's type of HDR without the potential alignment/ghosting issues. If only I had taken one more foreground and/or sky row, with better focus, ...
When I have nothing better to do, I may yet do a full HDR or Enfuse/TuFuse version of this scene based on the available files, but I don't expect to gain too much. I'll probably increase my chance of a really nice image by reshooting the scene with the lessons learned stored in the back of my head. Afterall, it probably won't be Friday the 13th again when the clouds cooperate next time and I go back there, so why not risk it?