Traces cells through 3D electron microscopy images with help from Mechanical Turk users. See connectomics.

This is a new program and not fully written yet but it can be used.

A "data" directory with an example stack of images is included with the code for demonstration.

Install python 2.7 with necessary modules. Install s3cmd

Use this command to configure s3cmd:

s3cmd --configure

Linux example:

cd dseg

python dseg.py data /home/rgiuly/output/test4 --zprocess --submit --sigma=4 --level=0.5 --access_key=YOURACCESSKEY --secret_key=YOURSECRETKEY --init  --seeds=[[473,44,10],[425,465,10]]

• --zprocess Run process for collecting decisions from users.
• --submit Submit decisions to Mechanical Turk.
• --sigma=4 Amount of blur before watershed.
• --level=0.5 Watershed level.
• --access_key=X Set to your access key from Mechanical Turk.
• --secret_key=X Set to your secret key from Mechanical Turk.
• --init Do initialization. This make take some time from a large volume.
• --seeds=[[473,44,10],[425,465,10]] Seeds where each cell should start in the form [X,Y,Z]. You can use IMOD to get the numbers.

To initially start the process from the beginning, use "--init". This will create initial superpixels (which takes a long time) and then it will start collecting decisions from users.

python dseg.py data /home/rgiuly/output/test4 --zprocess --submit --sigma=4 --level=0.5 --access_key=X --secret_key=X --init --seeds=[[473,44,10],[425,465,10]]

You can add parameters such as this. New regions will be sent to the database as they are discovered by users.

--send_regions_to_database --dataset_id=10821524 --model_id=2000

You can abort the process when results are being collected from users and pick up later. To continue collecting, leave out --init like this:

python dseg.py data /home/rgiuly/output/test4 --zprocess --submit --sigma=4 --level=0.5 --access_key=X --secret_key=X --seeds=[[473,44,10],[425,465,10]]

If you want to delete results collected from users so far and restart collection, use the --restart option like this:

python dseg.py data /home/rgiuly/output/test4 --zprocess --submit --sigma=4 --level=0.5 --access_key=X --secret_key=X --restart --seeds=[[473,44,10],[425,465,10]]

You can manually exclude nodes from the output with --delete, for example: --delete=['116_26','116_14','115_47']

Creating a qualification set that will be used to train users:

python dseg.py data /home/rgiuly/output/test4 --zqual --answers=~/answers1.txt --sigma=4 --level=0.5 --access_key=X --secret_key=X

Send superpixel regions checked by users to model in SLASH portal (galle.crbs.ucsd.edu) example:

python dseg.py data /home/rgiuly/output/test4 --send_regions_to_database --dataset_id=10821524 --model_id=2000

This approves submitted tasks, which allows Mechanical Turk users to be paid.

dseg.py data /home/rgiuly/output/test4 --approve_all --access_key=X --secret_key=X

Note: Suggested preprocessing of data, assuming you have a tif stack. This is for both qualification and processing. We use imagemagick to perform histogram equalization: cd to image folder. For example:

ls *.tif | xargs --verbose -I XXXX convert XXXX -negate -equalize -depth 8 contrast/first4/contrastXXXX

DP2: Distributed 3D Image Segmentation Using Micro-labor Workforce