Issues and annoyances with trakem2 working on a gigapixel montage

Posted by Stoyan Pavlov on
URL: http://imagej.273.s1.nabble.com/Issues-and-annoyances-with-trakem2-working-on-a-gigapixel-montage-tp5018056.html

Dear friends and colleagues,
First I want to emphasize, how astonishing and useful I find the trakem2
package/plugin! It is really great and I am really thankful to the
developers for puting all this effort into creating this suite for working
with large image datasets.
I use it mainly for creating/stitching/blending , processing and exporting
of gigapixel montages of microscopic images into a readable software
independent image pyramid.
Recently I stumbled upon several issues (I don't know whether they are bugs
or just due to limitation in the software) for which I couldn't find any
mention and/or solution in the documentation and the forum.

First I will describe what I am doing and next I will list the issues that
bother me.
I am creating a single layer montage with dimensions 63698 x 57218 px from
a set of 3111 images (each 1388x1040). The images are acquired via
axiovision and exported as single tiles (24bit RGB in either jpg or tif
format). Before importing the images I align them using the MIST plugin and
use the global positions generated from it to import them into trakem2 via
text file. The resulting montage is exactly as expected and have no remarks
about it.
1) I don't know if its true but it seems to me that if I use jpg as format
for the source images they are imported faster into trakem2. However once
the import is finished the software returns errors about generating mipmaps
(see attached log file) and it continuously tries to regenerate them (which
I guess keeps CPU quite busy). This happens regardless mipmaps are enabled
or disabled in Display>Properties, and is independent from the number of
threads assigned to mipmap generation. If I close the project and Fiji
(even after system restart) and then I open them again the issue returns.
This behaviour is not oberved if my source images are in tif format.
The other issues are related to the export of a flat image (here the file
format of the source images does not matter):

2) When I am trying to export to a tiled format (Export> Make flat
image...>Save for web (CATMAID)) regardless of the image format I choose
for the export I cannot change the size of the exported tiles to other than
256x256. The dialogue allows me to change it, but when the process finishes
in the folder there are only 256x256 tile images. While this is not a huge
problem, still there is difference between managing some 80000 images or
40000.

3) Also when I am trying to export tiled format  via Export> Make flat
image...>Save for web (CATMAID), if the patches are deselected only a
portion of the image is exported . Regardless of the format of the source
images and of the exported images there are only 63 rows by 63 columns of
256x256 tiles at zoom level 0 in the folder with the exported images. All
metadata in the json file is correct and there are no errors reported by
the System log or the Fiji Log. If the patches are selected with Select All
(Ctrl + A) the image is exported in its entirety, however here comes the
next not as much as an issue but slight discomfort.

4) The image is always exported to a quadrangular array consisting (for the
reasons already described) of 256 rows and 256 columns of 256x256 tiles at
zoom level 0. That is the image dimensions (63698 x 57218 px) are expanded
to 65536x65536 and filled with black tiles. This of course take up
additional space and need to be cropped away before uploading the image to
our servers. I understand that this is probably because of limitations to
the algorithm but still if there is some workaround it would have been
better.

5) I thought that this problems (2-4) might be due to the unequal
dimensions of the original source images and I tried to work around them by
creating a sibling project with retiled images. Here I had another
unexpected result - the image was exported but the larger dimension was cut
by about 6000 pxs from the right side  and than restored to the original
size by adding black pixels , which basically cuts away the right portion
of my image (see attached screenshot) .

The workstation that I use has the following parameters:
 CPU:  Intel Xeon CPU E5-2643 @3.3 GHz (2 x 8 core processors)
RAM: 64 GB
OS: Windows 7 64bit
In an attachment I added the ImageJ-properties output, portion of the
trakem2 log with the errors generating the mipmaps, and a screenshot of the
output of the sibling project with retiled images.

I hope someone will find a way to correct these problems, because surely
they are not fatal and there are workarounds that we use, but it is
certainly annoying.

Best regards,
Stoyan

---
Dr. Stoyan P. Pavlov, MD, PhD
Departament of Anatomy, Histology and Embryology
Medical University "Prof. Dr. Paraskev Stoyanov", Varna
Prof. Marin Drinov Str.55
9002 Varna
 Bulgaria
Tel: +359 (0) 52 - 677 - 052
e-mail: [hidden email]
           [hidden email]

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html