Skip to content

Commit

Permalink
unique class double annotate case, also check for previous handling
Browse files Browse the repository at this point in the history 
switch dup_coord order for final mask plotting of unresolved total_total cases
  • Loading branch information
@HaleySchuhl
HaleySchuhl committed Aug 13, 2024
1 parent fbf9fc2 commit daf2e8e
Showing 1 changed file with 62 additions and 61 deletions.
123 changes: 62 additions & 61 deletions plantcv/annotate/classes.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -268,69 +268,70 @@ def correct_mask(self, bin_img):
debug_img, final_mask, object_id_count = _draw_resolved(debug_img, final_mask, labeled_mask_all,
mask_pixel_value, object_id_count)
if associated_count > 1:
# Object annotated more than once so find all associated annotations
associated_coords = np.where(masked_image2 == mask_pixel_value)
associated_coords = tuple(zip(*associated_coords))
coord_labels = []
# Find all class labels for each annotation
for dup_coord in associated_coords:
# Flip x & y for numpy, and find the associated class label with each coordinate
coord_class_label = [k for k, v in self.coords.items() if (dup_coord[1], dup_coord[0]) in v]
coord_labels.append(coord_class_label)
print(coord_labels)
# Is there more than one class label associated with the given object?
re = np.unique(coord_labels)
if len(re) == 1:
# Labels are duplicated
# Draw the ghost of objects removed
debug_img_duplicates = np.where(labeled_mask_all == mask_pixel_value, (255), debug_img_duplicates)
if mask_pixel_value not in added_obj_labels:
# Fill in the duplicate object in the labeled mask, replace with pixel annotations
final_mask = np.where(labeled_mask_all == mask_pixel_value, (0), final_mask)
added_obj_labels.append(mask_pixel_value)
for dup_coord in associated_coords:
final_mask[dup_coord[1], dup_coord[0]] = object_id_count
analysis_labels.append(names)
# Add a thicker pixel where unresolved annotation to the debug img
cv2.circle(debug_img, (dup_coord[1], dup_coord[0]), radius=params.line_thickness, color=(object_id_count), thickness=-1)
# Add debug label annotations later
debug_coords.append((dup_coord[1], dup_coord[0]))
debug_labels.append(str(object_id_count))
# Increment object count up so each pixel drawn in labeled mask is unique
object_id_count += 1
print(np.unique(final_mask))
if len(re) > 1: # and count of each label = 1 (count is all "1"s??)
# More than one class label associated with a given object
splitup = []
# Split on "_" in case something has already been combined
for lbl in re:
list_lbl = lbl[0].split("_")
splitup.append(list_lbl)
# Flatten list of labels
flat = np.concatenate(splitup)
# Grab each unique label from the list
unique_lbls, lbl_counts = np.unique(flat, return_counts=True)
# Is there duplication within each class label for the given object?
if np.all(lbl_counts == 1):
# If no, Concat with "_" delimiter
concat_lbl = "_".join(list(unique_lbls))
# Adding the object
added_obj_labels.append(object_id_count)
analysis_labels.append(concat_lbl)
# Add debug label annotations later
debug_coords.append((x, y))
debug_labels.append(text)
# Draw on labeled mask and debug img
debug_img, final_mask, object_id_count = _draw_resolved(
debug_img, final_mask, labeled_mask_all, mask_pixel_value, object_id_count)
else:
# Has this object been handled already?
if mask_pixel_value not in added_obj_labels:
# Object annotated more than once so find all associated annotations
associated_coords = np.where(masked_image2 == mask_pixel_value)
associated_coords = tuple(zip(*associated_coords))
coord_labels = []
# Find all class labels for each annotation
for dup_coord in associated_coords:
# Flip x & y for numpy, and find the associated class label with each coordinate
coord_class_label = [k for k, v in self.coords.items() if (dup_coord[1], dup_coord[0]) in v]
coord_labels.append(coord_class_label)
print(coord_labels)
# Is there more than one class label associated with the given object?
re = np.unique(coord_labels)
if len(re) == 1:
# Labels are duplicated e.g. "total", "total"
# Draw the ghost of objects removed
debug_img_duplicates = np.where(labeled_mask_all == mask_pixel_value, (255), debug_img_duplicates)
if mask_pixel_value not in added_obj_labels:
# Draw the ghost of objects removed
debug_img_duplicates = np.where(labeled_mask_all == mask_pixel_value, (255), debug_img_duplicates)
# Fill in the duplicate object in the labeled mask, replace with pixel annotations
final_mask = np.where(labeled_mask_all == mask_pixel_value, (0), final_mask)
# If there are duplication in labels (e.g. [['total'], ['total']] then add to list)
dupes = [x for n, x in enumerate(coord_labels) if x in coord_labels[:n]]
added_obj_labels.append(mask_pixel_value)
for dup_coord in associated_coords:
final_mask[dup_coord] = object_id_count
analysis_labels.append(names)
# Add a thicker pixel where unresolved annotation to the debug img
cv2.circle(debug_img, (dup_coord[1], dup_coord[0]), radius=params.line_thickness, color=(object_id_count), thickness=-1)
# Add debug label annotations later
debug_coords.append((dup_coord[1], dup_coord[0]))
debug_labels.append(str(object_id_count))
# Increment object count up so each pixel drawn in labeled mask is unique
object_id_count += 1
print(np.unique(final_mask))
if len(re) > 1: # and count of each label = 1 (count is all "1"s??)
# More than one class label associated with a given object
splitup = []
# Split on "_" in case something has already been combined
for lbl in re:
list_lbl = lbl[0].split("_")
splitup.append(list_lbl)
# Flatten list of labels
flat = np.concatenate(splitup)
# Grab each unique label from the list
unique_lbls, lbl_counts = np.unique(flat, return_counts=True)
# Is there duplication within each class label for the given object?
if np.all(lbl_counts == 1):
if mask_pixel_value not in added_obj_labels:
# If no, Concat with "_" delimiter
concat_lbl = "_".join(list(unique_lbls))
# Adding the object
added_obj_labels.append(object_id_count)
analysis_labels.append(concat_lbl)
# Draw on labeled mask and debug img
debug_img, final_mask, object_id_count = _draw_resolved(
debug_img, final_mask, labeled_mask_all, mask_pixel_value, object_id_count)
else:
if mask_pixel_value not in added_obj_labels:
added_obj_labels.append(mask_pixel_value)
# Draw the ghost of objects removed
debug_img_duplicates = np.where(labeled_mask_all == mask_pixel_value, (255), debug_img_duplicates)
# Fill in the duplicate object in the labeled mask, replace with pixel annotations
final_mask = np.where(labeled_mask_all == mask_pixel_value, (0), final_mask)
# If there are duplication in labels (e.g. [['total'], ['total']] then add to list)
dupes = [x for n, x in enumerate(coord_labels) if x in coord_labels[:n]]

# Combine and colorize the debug image
# Dilate duplicate objs and subtract the object itself to leave just a halo around
Expand Down Expand Up @@ -419,7 +420,7 @@ def _draw_unresolved_object(debug_img, final_mask, obj_number, coord):
"""
(x, y) = coord
# Add a pixel where unresolved annotation to the mask
final_mask[y,x] = obj_number
final_mask[y, x] = obj_number
# Add a thicker pixel where unresolved annotation to the debug img
cv2.circle(debug_img, (x, y), radius=params.line_thickness, color=(obj_number), thickness=-1)
# Increment ID number up by one
Expand Down

0 comments on commit daf2e8e

Please sign in to comment.