first draft of BoxObjects #278
Conversation
|
I put together a small test pipeline using the 1st image in the example dataset. IDPrimary finds 296 objects. BoundingBoxes using 1st or overlap both make 296 objects. I used workspace viewer to confirm that generally the objects are mapping to the correct numbers (1st in green, Overlap in pink, IDPrimary in black) However, the final object made with 1st is almost all nan measurements. When I zoom in to the final object (296), workspace viewer shows object outlines that should have measurements for both box types. Also, in the object.csv's, I would expect the |
|
After digging into the first issue, I realized I was overwriting the ALLOW OVERLAP objects with the default non-overlapping object ("segmented") included in CellProfiler, I fixed this issue so now the objects are truly overlapping. I was unable to replicate the second error raised by Erin ("the final object made with 1st is almost all nan measurements") so I could not track down the cause. One issue raised after these changes is that the Measure modules do not operate properly using the truly overlapping objects. Going forward, I will try to crop out single images for each cell and process those independently, and/or will discuss with Beth about updating the Measure modules to allow for overlapping objects ("ijv" instead of "segmented") |
| def get_folder_name(self, workspace, filename_values): | ||
| """Get original folder name based on user input""" | ||
| if self.file_name_method == FN_FROM_IMAGE: | ||
| filename_image = [f for f in filename_values if f.startswith("r")][0] |
There was a problem hiding this comment.
Seems like this is hardcoded to match specific file name structuring? I tested it on the demo pipeline images and it crashes here.
| segmented[y_min:y_max, x_min:x_max] = object_id | ||
|
|
||
| cropped_mask = input_label[y_min:y_max, x_min:x_max] == object_id | ||
| mask_save_filename = f"Object{image_name}_{self.output_object_name.value}_{object_id}.tiff" |
There was a problem hiding this comment.
image_name should not be a part of the mask filename as objects are independent of images
Cellprofiler plugin with creates bounding box objects from the previously segmented objects