Yes, it could be done with that amount of images with a good computer : windows or mac, 64 bits needed.

I advises you at least 12 GB RAM, reduce the memory allocated to rendering so that everything is just for detection. Do the detection and then only go back to the general memory setting to raise again this value when you are done with the detection.

AlexandreJ wrote:

Do the detection and then only go back to the general memory setting to raise again this value when you are done with the detection.

Alex wouldn't it be better if autopano knew what it is doing & switch the memory allocation automatically?

digipano wrote:

Alex wouldn't it be better if autopano knew what it is doing & switch the memory allocation automatically?

Sure, we tried to do that ... not that easy.

Guilhem wrote:

Have you tried to put 1000mm focal inside APG ? Its the same work as for reproduction, isn't it ?

1000mm works well for art repro work since its almost flat, but aerial images are not perfectly flat they have 3d objects & landmark too. More over they curve & don't form a straight line of the straight flight path I guess due to spherical surface of earth.

1. You can use the settings->optimization tab-> multiple viewpoints checkbox .  I have used the 1000 mm trick in other stitchers, but APG has yielded better results.  I will need to try the very large focal length to observe any improvements.

2. You can create another folder of reduced resolution images to quickly try different methods.  Then at full size when you have figured it out.
    a.  Stitch overall vertical imagery in pieces.  Piece these together in Photoshop using transform warp, gradient mask / photomerge.   These smaller groups will stitch as a more uniform plane with less alignment problems.  It would be great to have some type of tool to connect these smaller groups together.  Photoshop photomerge can cause problems.  I often merge the different large images using a gradient mask to blend.  Sometimes photomerge actually works better.  It varies from merge to merge of images.  ( This would actually be a big help for a large project that I am currently working on. )

3. When photographs deviate from the Nadir / Normal or are at different distances, you can correct the photo ( orthorectify the image and scale ) before including it in a stitch using photoshop.  I have found images captured at different distances cause the most problems during alignment.

4. Calculating flight lines for near vertical imager ( vertical imagery is technically less than 4 degrees from the nadir ) you do need to more overlap than in a normal panorama in order to deal with objects that are tall.   Usually your overlap between photos on a particular flight line will be 60% and the sidelap between photos of different flight lines will be around 20%.  This is especially important if you are flying at a lower elevation such as 250 meters to 150 meters from the ground.

5.  Also, it is often best to capture near vertical imagery at solar noon to minimize shadows.

The blend times for rendering are reduced substantially for groups of 100 to 300 images when you increase the amount of memory from 8GB to 16GB or 24GB.  I am wondering if Alexandre and Lionel have looked at disk address traces to determine if IO is largely sequential and not random IO.  SSD would give you the largest performance increases for random IO.

Last edited by Castillonis (2011-08-26 13:35:11)