- OpenDiamond
- Intel IPP (tested with 2011.13.367)
- OpenDiamond
- 64-bit environment
- OpenDiamond
- virtualenv
- OpenCV-Python
- BLAS library and headers
- LAPACK library and headers
- Build the autosegmenter. Be sure to set up your
environment for IPP using
compilervars.sh
. - Use make-filter.py to build the Gemini filter.
- Create and populate virtualenv. Use
--system-site-packages
, since OpenDiamond can't be installed with pip. - Configure and run.
dermshare.py
is always needed, butdermshare-ws.py
is only necessary if the mobile device link is enabled withWEBSOCKET_BASEURL
.
To avoid conflicts with older versions of NumPy and Six packaged by your Linux distribution, it is best to (re-)install NumPy v1.9.1 and Six v1.9.0 with pip after installing OpenCV-Python.
fil_gemini
must install dependencies into a virtualenv in /tmp the first
time it runs. To avoid long delays in the first search, the virtualenv can
be prepopulated by running fil_gemini
without arguments on the Diamond
servers. To avoid additional delays whenever a Diamond server is rebooted
or /tmp is cleared, you can globally install the necessary Python libraries:
pip install -r gemini/requirements.txt
An example scope server is provided in simplescope. The scope server shipped with OpenDiamond will not work because it doesn't integrate with Diamond web apps.
An example dataretriever is provided in dataretriever. It
includes support for serving segmentation images and pathology reports
in addition to the Diamond objects and scope lists. The diamond_store
dataretriever shipped with OpenDiamond will probably also work, but has not
been tested.
The DermShare dataretriever expects a directory tree with the following layout:
DERMSHARE_SCOPE_PATH
<scope1>
<name1>-image.{jpg|png|...}
<name1>-segm.png
<name1>-attr.yaml
<name1>-path.html
<name2>-image.{jpg|png|...}
...
<scope2>
...
The segmentation image must be a PNG the same size as the image, with the pixels corresponding to the lesion marked in red and the remaining pixels transparent.
The attributes file, if present, contains only one attribute, the diagnosis code as listed in search.js:
diagnosis: bcc
The path file, if present, contains the pathology report to be shown in DermShare.