Preprocessing the ground truth annotations¶
Offset correction and concatenating the annotation files¶
We use ground truth files in .ndjson format with the particle centroid coordinates mentioned with respect to the origin situated at the top left front corner of the tomogram.
For TomoTwin dataset¶
We noticed that the ground truth annotations had an offset. This was corrected by running the tomotwin ground truth preprocessing script which can be run as follows:
python accessories/preprocess_ground_truth/generate_ndjson_ground_truth_tomotwin.py <parent_dir> <x> <y> <z>
where, parent_dir is the directory of a round of simulated tomograms, and x, y and z denote the size of the tomogram along the three axes.
For CZI datasets¶
We noticed that the ground truth annotations did not have any offset. The coordinate files were concatenated to a single file by czi ground truth preprocessing script which can be run as follows:
where <parent_dir> corresponds to the annotations directory associated with each tomogram.