I was using the tracking feature of Ilastik (inputs: Raw Data, Pixel Prediction Map) and noticed that the export results (Export Source: Plugin, CSV-Table) from the Batch Processing mode is wrong, while export results from the GUI export button under 6. Tracking Results Export menu of the same set of input data is correct. It seems there’s a bug in batch processing/headless mode using the plugin/CSV-Table option. I have attached necessary files and related input/output that might be useful to the debugging. This is a tracking project with only 3 time points (frames). There’s only one set of input images in the attached Ilastik project file:
Ilastik project file: tracking_test.ilp.zip (compressed to reduce size)
Export csv-table results:
from GUI export: r03c24f01p01-ch1sk1fk1fl1_raw_data_CSV-Table_exportfromGUI.csv
from batch export: r03c24f01p01-ch1sk1fk1fl1_raw_data_CSV-Table_exportfromBatchExport.csv
I have visualized the export results, where object_center_0 and object_center_1 are used as X and Y axes and objects size reflect their object_area value. Circles are colored by the three time points. They should roughly match the original raw image or probability map in terms of object locations and sizes.
As we can see the results from GUI export (GUIexportSpotfire.png) matches original images flawlessly, but the result from batch export (BatchExportSpotfire.png) does not. It seems that the results from the batch export contains mistakes in both locations and sizes (many object locations are pushed to the right edge), and for those with seemly correct locations, their sizes are off. I also attached the command line I used, however, the bug can be reproduced with the GUI batch export.
Thank you so much in advance for looking into this issue. Looking forward to hear back from you!
first of all, welcome to the image.sc forum!
And then, thank you very much for your detailed report and for providing all the data. I can reproduce the issue (and am of course puzzled by it). We’ll investigate it!