Gigapan: Lindaunis Brücke

Bridge near the village of Lindaunis in Schleswig-Holstein, Germany (Wiki)

Gigapan: View from Pinnacles @ Duxton 2

Another viewpoint.

Gigapan: View from Pinnacles @ Duxton

Gigapan: Pearl Bank Apartments

Seen from Pinnacle@Duxton rooftop garden.

Gigapan: Construction Site

When I created panoramas previously I usually dont exceed 100 MegaPixel because it is good enough for (very) large prints and I rather have a perfect 100 MP version without seams and changing light conditions than a 1GPixel version with all these faults. Until some time back my hardware was also limited and large stitches were not an option. Now any decent desktop PC can easily be equipped with 12 to 16GB RAM and with a Windows or Ubuntu 64 bit version on a recent CPU, you get fast rendering times.

I just made a 550 MPixel (or 0.55 GigaPixel) created from 60 images shot with Canon 5D MK II and 7-200 L IS.

Gigapan: Maximilian Strasse Speyer

Gigapan: Wachtenburg Wachenheim

(32 images at 200mm, 5D II, L IS 70-200, NN5)

Gigapixel Record Hunt

2 years back there were only few images on gigapan.org with a serious "big" resolution, now with the production release EPIC devices more highres panoramas are out there.
A bit longer breathtaking new super records of high resolution images come up every now and then.

• First there was Bryce Canyon National Park with 1.09 GP breaking the 1 GP barrier in 2003
  
  
  
• Then came the Harlem 13 GP Project (2006)
  
  
  
• Then Yosemite in 17 GP
  
  
  
• Then Paris and Dresden followed with 26 GP
  
  
  
• Dubai in 45 GP (April 2010)
  
  
  
• and now Budapest in 70 GP !!!
  

Conclusion: We will see more of this kind coming up, and it is amazing at what detail level you can "zoom around" in these images. I recomend reading the technical background of the above projects, as you need serious project preparation, support (maybe some $$$) and serious CPU power. Guess I will stick to the "less-than-10 GP" department.

Remarks:

• You find more gigapixel project if you research, some of them based on film (Gigapxl Project). Unfortunately some of the early projects are not online anymore.

GIGAPAN Chinese Garden

Another GIGAPAN made with the modified gigapan systems head and the Panasonic Lumix FZ28. 96 images stitched with PTGui to a 14.200 x 17.300 pixel image (245MP). Not sure if it reveals much more details than 4 Canon 1Ds images stitched together. Maybe I will compare for this sake.

See the complete GIGAPAN here (link)

Lumix FZ28 goes gigapan

Managed to mount a Panasonic Lumix FZ28 on the gigapan head with the help of a little homemade support bracket from the DIY shop. Not the final version, but a start.

GIGAPAN Tips and Tricks (Part 2)

This topic does not apply specifically to images created with the Gigapan system, but to all stitching practice in general. The gigapan gives you more control over the process than for example handheld shot panoramas or using a panohead without dents.

I am a perfectionist when it comes to stitching images. I dislike (creating or printing) panoramas where you can see obvious seams or moving subjects that show up half (usually referred to as ghosts).

Preconditions:
a) All images are shot with same camera settings (whitebalance, aperture, speed). Optional usage of RAW.
b) The more overlap the better (to some extend only!). I recommend somewhere 30..45%. This you set during the Gigapan Setup (FOV setting) or manually while shooting with non-gigapan gear.
c) The more images (or the higher the cameras resolution) you want to stitch the longer it will take, resp. require more memory in order to get results in a realistic timeframe. I am not sure what are the limitations of running Gigapan Stitcher under Windows 32.
d) Set the "Kepp projected images" option in the Gigapan Stitcher Options screen (see Img.1)

Img.1

Retouch/Post Processing:
With plenty of overlap I created the 2 images (detail crop img.2) of the same spot with the scene itself changing. The stitcher cannot take care of moving subjects like people or cars. You can se the result in img. 3 (upper part). To remove the ghost appearance you load the warped images with the overlapping scenes from the data/projected folder (applies only to gigapan stitcher) into Photoshop. If you use PTGui save the resulting panorama file either as image with all layers or save the layers as separate files. Now you copy and paste parts of the scene to repair from the warped images into the stitched "big" file (use 50% transparency to match the non-moving subjects like flagpole, etc.) and use the layer tools to reveal or hide the parts to repair. I find this as a very efficient and fast way to repair the ghosts and its way better than cloning and stamping subjects, background patterns or copy-n-paste-mirror-resize bodyparts. See he result in img.3 or goto gigapan.org (link) to see the fullsize version.

Tips for Photoshop:
(Big panoramas require lots of memory for a decent working speed.)
- Use Windows 64 with lot of real memory (not always an option due to availability)
- Terminate any other non-essential programs or services.
- Switch off the copy to clipboard function (see PS options/settings).
- Assign 75..80% of memory to PS (see PS options/settings).
- Reduce the PS history to 2 or max 3, worst case to 1 (see PS options/settings).
- Do a regular EDIT/PURGE/ALL to relief memory (seems to work for me).
- Be patient.

Img.2

Img.3

New Stitching application

During the last few years a couple of image stitching applications appeared on the scene and most of them matured well, means upgrades with faster performance, more functionality, better results and so on. Just to name a few (at leats these ones I ever used):

Panorama Factory (link)
Autopano Pro (link)
PTGui (link)
Hugin (link)

Today I prefer to work with PTGui with a truckload of manual functionality, as well superb results in auto-mode.

Btw, the Gigapan system is "shipped" with a stitcher as well. (link)

Microsoft Research just launched another new stitcher (ICE, Image Composite Editor) in the neighborhood of Deepzoom Composer and co. I just gave it a try and the results are pretty good. I will continue to test.

Some more info (link)
Download (link)

GIGAPAN Tips and Tricks (Part 1)

• Check if the camera fits into the GIGAPAN camera bracket, the Panasonic LX3 almost doesnt fit. Check the list of tested cameras here (link)
  
• Use a camera that allows manual override, at least with a exposure lock, otherwise dont waste your time. Results are disappointing (due to the camera producing different exposed pictures, not the fault of the gigapan system).
• You might want to look out for a camera that supports RAW to get some extra control on color, exposure and detail(?).
  
• Best result with these camera settings: manual exposure, manual or fix whitebalance (aka sun, cloud,..), fix focus.
  
• Dont use digital zoom ! It is not going to increase the resolution ! Switch it off.
• The more overlap you have between the images, the higher the chance you can patchwork (in PS or GIMP) some of the moving and most probably half objects.

First GIGAPAN

Finally I get my hands dirty with the GIGAPAN head. I brought it from Singapore (after it was shipped from US) back to Germany to spend some time experimenting with it. I share some first hands-on experience with it. Ultimately I will compare the GIGAPAN setup with stitched images from 1Ds and Mamyia ZD and a spherical scanner (even complete different league of equipment!). I will look into usability in the field.

Technical Data:
- Camera New Panasonic LX3 (10 MP, jpeg high). Manual settings, fix white balance. Manual focus.
- 25 images in 3 rows and 5 columns
- Setup and shooting in less than 5min. Finding a spot 30min.
- Stitched with Gigapan Stitcher 12.619 x 4.687 pixel (59.1MP) in roughly 10min.
- Photoshop to crop and slight curve adjustments.

Conclusion:
- The setup of the head is fast. The one-time (per camera) setup you better do at home and get comfortable with the settings of both camera and GIGAPAN
head.
- The total shooting time depends on the camera, (How fast can it store the image data ? Does it run out of buffer after 10 or so shots ?), the size of FOV and the area you want to cover. The more you can zoom in (with the lens), the smaller the FOV the more pictures you have to take.
You should experiment with the settings, especially with the break in seconds between the shots before you go out and start shooting.
I tried with the (slower) BenQ camera and the GIGAPAN head was triggering faster than the camera can digest, ending up single shots got lost.

Options:
- Shooting in RAW, would give a bit more control later on color temperature, maybe a slightly (?) higher level of detail. But most cameras cannot shoot continously in RAW without the need to clear the buffer to the memory after a number of shots (the same time you cant do anything but wait). Note there is no "data" connection between head and camera. The head does nothing but moving the camera and triggering mechanically.
- Stitch with PTGUI to achieve more flexibility, eg. stitch to PSD format with individual layers, provided you have enough memory and CPU power.
I will upload the result in original size to gigapan.org once I am back to my office with fast broadband connection. So this is a downsized preview only.
I will add more test results and feedback as I proceed testing. Stay tuned.

Gigapan out of the box

In less than a week after ordering my beta-tester set of the Charmed Labs (link) Gigapan head arrived. Under the Gigpan-Systems (link) label they started the production of this automatic panohead for consumer sized cameras. The results are currently on an own website (link).

My first impression:
smaller and lighter than expected
no prototype look-and-feel
easy to setup and use,
fun,looking forward to experiment with it

I will use it with a Panasonic LX3.
You need to wait for my first results and in-deep test, since I am just packing for Germany for 3 weeks.

The picture show the unpacking ceremony and the test setup with el-cheapo Benq cam.

New Panorama Engine on the way

For quite some time panorama photographers already working on multirow panoramas with highres equipment and/or large computing setups to process the huge amount of images/data (check out the Harlem 13 Gigapixel Panorama at link). If you do a manual multirow you are challenged, if your go for more than 3 rows because I find it not easy to manually adjust the (panoramic) head accordingly. There are a number of automated solutions out there, most of them in beta state or homebrew stuff. If you followed the Gigapan (link) project you saw their prototype more than one year ago. Finally they ship the first batch of panoheads. Since I am registred as beta user, I ordered (its not free) my tripod head today. Looking forward to work with it. I will keep you updated.