The conference will run a stitching challenge to evaluate state-of-the-art algorithms. Stitching to is a necessary step for most microscopes supporting automated acquisition of image tiles (fields of views).
Participants of the stitching challenge will be able to download image tiles and metadata information about the acquisition and the arrangement of each tile from the NIST web site (see the URL links below). They will then run their corresponding translation-based stitching algorithm on the provided datasets. The results of any stitching algorithm should be emailed to the challenge chair in a csv file format. The csv file should contain the global image tile locations with respect to the upper left tile (sample csv file is provided). In addition, participants will fill out an xml spread sheet with information about the stitching algorithm, and software and hardware environments used to execute the algorithm. The submission and notification deadlines follow the conference paper deadlines.
The challenge dataset is a grid of nxn image tiles of stem cell colonies. The images were acquired with 10% overlap. There are three levels of difficulty ranging from easy (level 1), intermediate (level 2) and hard (level 3). Level 1 is acquired in feature rich phase contrast imaging modality with grid size 10x10, Level 2 is acquired in Cy5 (where signal exist only on top of colonies and mainly background noise fills the rest of the image) with grid size 10x10 and level 3 is acquired in Cy5 with a larger grid size of 24x23. Participants can send results for any of the three levels and the results will be evaluated at each level separately.
All grid tiling starts in the upper left corner and moves horizontally (x or column direction). The shift between rows is done by combing, meaning the next tile acquired after the 10th right tile on row i is the first tile on the left of row i+1.
The naming convention of the files follows the one done by MicroManager in the format: file_name_r{rrr}_c{ccc}.tif where r{rrr} is the row number and c{ccc} is the column number of the tile in the 10x10 grid. For example: img_Cy5_r009_c006.tif is the image acquired using Cy5 and located on row 9, column 6 on the 10x10 grid.
The results of any stitching algorithm will be submitted to the challenge chair in a csv file containing the image name and its global position within the stitching image. The global position csv file contains one image position per line formatted as such: “<image name>, <x coordinate>, <y coordinate>”. For example: “img_Cy5_r001_c001.tif, 0, 40”. An example of a global position file, “global-positions.txt”, and a submission file, "BII_Stitching_Submission_Example.xml", are available for reference (see the URL links below).
Level1_Image_Tiles.zip ~ 195 MB
Level2_Image_Tiles.zip ~ 119 MB
Level3_Image_Tiles.zip ~ 665 MB
BII_Stitching_Submission_Example.xml ~ 2 KB
The challenge committee will use the global locations in a csv file to assemble the resulting stitched image. The accuracy of the stitching algorithm will be determined by comparing two measures: (1) colony centroid location and (2) colony area obtained from the stitched image and from the measured sample (reference value). The results would be reported at the conference.
National Institute of Standards and Technology
Joe Chalfoun
joe.chalfoun@nist.gov
Phone: 301.975.3354
September 15, 2015
Via email to the challenge chair