Working with Chunks

[IanManning]

Hi there folks,

A quick question regarding working with Chunks in PhotoScan. I've got a large set of photos ~550. The photos are taken with a UAV, and the total scene include 8 control points with average GPS locations.

My end goal is to produce both an Orthophoto mosaic and Digital Terrain model of the study area. I can run all the required processing with a Medium setting, however my computer runs out of memory when using high quality photos.  Small details are really important for the final product, so I'd really like to create a product at the highest settings.

After doing some reading on this forum, and other online PhotoScan resources, it looks like the best way to circumvent this memory issue is by processing my project into separate chunks. I've given it a shot, but unfortunately the final product was poor.

My questions are:

1. Is using chunks the best/only way to work-around the memory shortage warning?
2. If so, is there an optimal time to split the dataset into chunks? E.x before photo alignment, after dense cloud.

3. When is the best point in the workflow to merge chunks?


Thank you in advance,
Ian

[Alexey Pasumansky]

Hello Ian,

In case there's no way to avoid splitting, I suggest to use the same alignment results, duplicate the original chunk multiple times, use bounding box to reduce the reconstruction volume in the individual chunks, build dense cloud and mesh and then merge the results into single chunk for DEM/Orthophoto export.

There's a script sample that performs the given operation in the automatic way:
http://wiki.agisoft.com/wiki/Split_in_chunks.py

[IanManning]

Thank you Alexy,

This is exactly what I was looking for. I'll be sure to report back with my results. Thanks!

Best,
Ian

[Alexey Pasumansky]

Hello Ian,

By the way, please specify the amount of RAM installed on your machine and the processing stage that faces the memory limit (dense cloud generation or mesh reconstruction)?

[IanManning]

Sure thing Alexy

My PC has 16GB Ram.

In this particular case, I ran out of memory during the Dense Cloud generation processing stage.

Thanks,
Ian