[Qgis-user] Re: qgis and big raster catalogs

[dncarreira]

Hi there, and thanks for the replies! Those got me into the right
direction... even though there are a few snags in qgis (see end of message). 

What I ended up doing: 

1) create graticule using gdaltindex - this will serve only as a visual
guide to the user 
2) create a vrt catalog using gdalbuildvrt 
3) edit the vrt file manually to insert band statistics - so the vrt will
load fast in qgis 
4) create a mosaic with 25x bigger pixel resolution (2.5m) 
5) build overviews for the mosaic 
6) create a qgis project where the mosaic is visible until 1:9000 scale is
reached 
7) edit the qgis project manually to set the minimum visibile scale of the
vrt catalog to 1:9000 

This got me a pretty good performance. I could even load 2 sets of images
instead of just 1, with a total of 880 ecws. (Note that qgis gets hold of
~1GB ram in my pc.) 

Now there are 2 problems in QGIS that I detected or confirmed: 
1) if a vrt file does not have statistics, qgis hangs while loading the vrt
file. You have to add statistics to the vrt. To make it faster I create a
small vrt with a few ecws and built the statistics using gdalinfo -stats.
Then copied those statistics into the big vrt catalogs. 
2) QGIS will hang on opening the properties of a vrt catalog, I assume when
it's big enough... couldn't find the reason. So to change the properties of
these layers I had to save the project and edit manually. I suspect the
problem could be qgis is building the histogram... but could not find a way
to confirm. 

After this experiment I feel a somewhat simple requirement gets really
complicated, not because the software couldn't handle it but because there
are few loopholes (2) to solve in qgis in this particular area - big raster. 

Hope this is helpful to someone else, and that the devs pick this up too ;) 

Duarte

--
View this message in context: http://osgeo-org.1560.n6.nabble.com/qgis-and-big-raster-catalogs-tp4672179p4715613.html
Sent from the Quantum GIS - User mailing list archive at Nabble.com.