Commit
·
d7401be
1
Parent(s):
1cdfb96
add: sep visualization for 9:16 and 16:9
Browse files- .gitignore +2 -1
- crop_utils.py +172 -141
.gitignore
CHANGED
@@ -1 +1,2 @@
|
|
1 |
-
__pycache__
|
|
|
|
1 |
+
__pycache__
|
2 |
+
.DS_Store
|
crop_utils.py
CHANGED
@@ -175,7 +175,7 @@ def crop_and_draw_divisions(
|
|
175 |
Returns:
|
176 |
tuple: (cropped_image_16_9, image_with_lines, cropped_image_9_16)
|
177 |
"""
|
178 |
-
yolo_model =
|
179 |
# Calculate division width and boundaries
|
180 |
division_width = input_image.width / num_divisions
|
181 |
left_boundary = (left_division - 1) * division_width
|
@@ -187,7 +187,11 @@ def crop_and_draw_divisions(
|
|
187 |
)
|
188 |
|
189 |
# Run YOLO on the 9:16 crop to get person bbox
|
190 |
-
bbox =
|
|
|
|
|
|
|
|
|
191 |
x1, y1, x2, y2 = bbox
|
192 |
|
193 |
# Calculate top boundary with head margin
|
@@ -348,7 +352,7 @@ def find_persons_center(image):
|
|
348 |
int: x-coordinate of the center point of all persons
|
349 |
"""
|
350 |
# Detect persons (class 0 in COCO dataset)
|
351 |
-
results = model(image, classes=[0])
|
352 |
|
353 |
if not results or len(results[0].boxes) == 0:
|
354 |
# If no persons detected, return center of image
|
@@ -411,7 +415,7 @@ def create_layouts(image, left_division, right_division):
|
|
411 |
cutout_height = cutout_image.shape[0]
|
412 |
|
413 |
# 2. Run YOLO on cutout to get person bounding box and center
|
414 |
-
results = model(cutout_image, classes=[0])
|
415 |
|
416 |
# Default center if no detection
|
417 |
cutout_center_x = cutout_image.shape[1] // 2
|
@@ -578,9 +582,9 @@ def create_layouts(image, left_division, right_division):
|
|
578 |
return layout_crops, cutout_pil, cutout_16_9_pil, cutout_9_16_pil
|
579 |
|
580 |
|
581 |
-
def
|
582 |
"""
|
583 |
-
Create a visualization showing
|
584 |
Each crop region is outlined with a different color and labeled.
|
585 |
All crops are centered on the person's center point.
|
586 |
|
@@ -588,9 +592,10 @@ def draw_all_crops_on_original(image, left_division, right_division):
|
|
588 |
image: PIL Image
|
589 |
left_division: Left division index (1-20)
|
590 |
right_division: Right division index (1-20)
|
|
|
591 |
|
592 |
Returns:
|
593 |
-
PIL Image: Original image with
|
594 |
"""
|
595 |
# Convert PIL Image to cv2 format
|
596 |
if isinstance(image, Image.Image):
|
@@ -613,7 +618,7 @@ def draw_all_crops_on_original(image, left_division, right_division):
|
|
613 |
cutout_image = image_cv[:, left_boundary:right_boundary].copy()
|
614 |
|
615 |
# Get YOLO detections for person bounding box
|
616 |
-
results = model(cutout_image, classes=[0])
|
617 |
|
618 |
# Default values
|
619 |
cutout_center_x = cutout_image.shape[1] // 2
|
@@ -668,171 +673,184 @@ def draw_all_crops_on_original(image, left_division, right_division):
|
|
668 |
font_scale = 0.8
|
669 |
font_thickness = 2
|
670 |
|
671 |
-
#
|
672 |
-
|
673 |
-
|
674 |
-
|
675 |
-
|
676 |
-
|
677 |
-
|
678 |
-
|
679 |
-
|
680 |
-
|
681 |
-
|
682 |
-
|
683 |
-
|
684 |
-
|
685 |
-
|
686 |
-
|
687 |
-
|
|
|
688 |
|
689 |
-
#
|
690 |
-
|
691 |
-
|
|
|
|
|
|
|
692 |
|
693 |
-
|
694 |
-
|
695 |
-
|
696 |
-
if target_height_16_9 <= height:
|
697 |
-
# Calculate 20% of person height for top margin
|
698 |
-
top_margin = int(original_person_height * 0.2)
|
699 |
|
700 |
-
|
701 |
-
|
702 |
|
703 |
-
|
704 |
-
|
705 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
706 |
|
707 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
708 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
709 |
cv2.rectangle(
|
710 |
visualization,
|
711 |
-
(
|
712 |
-
(
|
713 |
-
colors["
|
714 |
thickness,
|
715 |
)
|
716 |
cv2.putText(
|
717 |
visualization,
|
718 |
-
"
|
719 |
-
(
|
720 |
font,
|
721 |
font_scale,
|
722 |
-
colors["
|
723 |
font_thickness,
|
724 |
)
|
725 |
|
726 |
-
#
|
727 |
-
|
728 |
-
|
729 |
-
|
730 |
-
|
731 |
-
x_start = max(
|
732 |
-
0,
|
733 |
-
min(
|
734 |
-
left_boundary + cutout_width - target_width_9_16,
|
735 |
-
original_center_x - target_width_9_16 // 2,
|
736 |
-
),
|
737 |
)
|
738 |
-
|
739 |
cv2.rectangle(
|
740 |
-
visualization,
|
|
|
|
|
|
|
|
|
741 |
)
|
742 |
cv2.putText(
|
743 |
visualization,
|
744 |
-
"
|
745 |
-
(
|
746 |
font,
|
747 |
font_scale,
|
748 |
-
colors["
|
749 |
font_thickness,
|
750 |
)
|
751 |
|
752 |
-
#
|
753 |
-
|
754 |
-
|
755 |
-
|
756 |
-
|
757 |
-
|
758 |
-
|
759 |
-
(
|
760 |
-
|
761 |
-
|
762 |
-
|
763 |
-
|
764 |
-
|
765 |
-
|
766 |
-
|
767 |
-
|
768 |
-
|
769 |
-
|
770 |
-
|
771 |
-
|
772 |
-
|
773 |
-
|
774 |
-
|
775 |
-
third_width = width // 3
|
776 |
-
third_left_x = max(
|
777 |
-
0, min(width - third_width, original_center_x - third_width // 2)
|
778 |
-
)
|
779 |
-
third_right_x = third_left_x + third_width
|
780 |
-
cv2.rectangle(
|
781 |
-
visualization,
|
782 |
-
(third_left_x, 0),
|
783 |
-
(third_right_x, height),
|
784 |
-
colors["third"],
|
785 |
-
thickness,
|
786 |
-
)
|
787 |
-
cv2.putText(
|
788 |
-
visualization,
|
789 |
-
"Third Width",
|
790 |
-
(third_left_x + 5, 120),
|
791 |
-
font,
|
792 |
-
font_scale,
|
793 |
-
colors["third"],
|
794 |
-
font_thickness,
|
795 |
-
)
|
796 |
-
|
797 |
-
# Two-thirds width layout
|
798 |
-
two_thirds_width = (width * 2) // 3
|
799 |
-
two_thirds_left_x = max(
|
800 |
-
0, min(width - two_thirds_width, original_center_x - two_thirds_width // 2)
|
801 |
-
)
|
802 |
-
two_thirds_right_x = two_thirds_left_x + two_thirds_width
|
803 |
-
cv2.rectangle(
|
804 |
-
visualization,
|
805 |
-
(two_thirds_left_x, 0),
|
806 |
-
(two_thirds_right_x, height),
|
807 |
-
colors["two_thirds"],
|
808 |
-
thickness,
|
809 |
-
)
|
810 |
-
cv2.putText(
|
811 |
-
visualization,
|
812 |
-
"Two-Thirds Width",
|
813 |
-
(two_thirds_left_x + 5, 150),
|
814 |
-
font,
|
815 |
-
font_scale,
|
816 |
-
colors["two_thirds"],
|
817 |
-
font_thickness,
|
818 |
-
)
|
819 |
|
820 |
-
#
|
821 |
center_radius = 8
|
822 |
cv2.circle(
|
823 |
visualization,
|
824 |
-
(original_center_x,
|
825 |
center_radius,
|
826 |
(255, 255, 255),
|
827 |
-1,
|
828 |
)
|
829 |
cv2.circle(
|
830 |
-
visualization,
|
|
|
|
|
|
|
|
|
831 |
)
|
832 |
cv2.putText(
|
833 |
visualization,
|
834 |
"Person Center",
|
835 |
-
(original_center_x + 10,
|
836 |
font,
|
837 |
font_scale,
|
838 |
(255, 255, 255),
|
@@ -852,8 +870,8 @@ def get_image_crop(cid=None, rsid=None, uid=None):
|
|
852 |
Returns:
|
853 |
gr.Gallery: Gallery of all generated images
|
854 |
"""
|
855 |
-
image_paths = get_sprite_firebase(cid, rsid, uid)
|
856 |
-
|
857 |
# Lists to store all images
|
858 |
all_images = []
|
859 |
all_captions = []
|
@@ -908,14 +926,27 @@ def get_image_crop(cid=None, rsid=None, uid=None):
|
|
908 |
mid_image, left_division, right_division
|
909 |
)
|
910 |
|
911 |
-
# Create the visualization with
|
912 |
-
|
913 |
-
mid_image, left_division, right_division
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
914 |
)
|
915 |
|
916 |
-
|
917 |
-
|
918 |
-
all_captions.append(f"All Crops Visualization {all_crops_visualization.size}")
|
919 |
|
920 |
# Add input and middle image to gallery
|
921 |
all_images.append(input_image)
|
|
|
175 |
Returns:
|
176 |
tuple: (cropped_image_16_9, image_with_lines, cropped_image_9_16)
|
177 |
"""
|
178 |
+
yolo_model = model
|
179 |
# Calculate division width and boundaries
|
180 |
division_width = input_image.width / num_divisions
|
181 |
left_boundary = (left_division - 1) * division_width
|
|
|
187 |
)
|
188 |
|
189 |
# Run YOLO on the 9:16 crop to get person bbox
|
190 |
+
bbox = (
|
191 |
+
yolo_model(cropped_image_9_16, classes=[0], conf=0.6)[0]
|
192 |
+
.boxes.xyxy.cpu()
|
193 |
+
.numpy()[0]
|
194 |
+
)
|
195 |
x1, y1, x2, y2 = bbox
|
196 |
|
197 |
# Calculate top boundary with head margin
|
|
|
352 |
int: x-coordinate of the center point of all persons
|
353 |
"""
|
354 |
# Detect persons (class 0 in COCO dataset)
|
355 |
+
results = model(image, classes=[0], conf=0.6)
|
356 |
|
357 |
if not results or len(results[0].boxes) == 0:
|
358 |
# If no persons detected, return center of image
|
|
|
415 |
cutout_height = cutout_image.shape[0]
|
416 |
|
417 |
# 2. Run YOLO on cutout to get person bounding box and center
|
418 |
+
results = model(cutout_image, classes=[0], conf=0.6)
|
419 |
|
420 |
# Default center if no detection
|
421 |
cutout_center_x = cutout_image.shape[1] // 2
|
|
|
582 |
return layout_crops, cutout_pil, cutout_16_9_pil, cutout_9_16_pil
|
583 |
|
584 |
|
585 |
+
def draw_crops_on_original(image, left_division, right_division, crop_types):
|
586 |
"""
|
587 |
+
Create a visualization showing selected crop regions overlaid on the original image.
|
588 |
Each crop region is outlined with a different color and labeled.
|
589 |
All crops are centered on the person's center point.
|
590 |
|
|
|
592 |
image: PIL Image
|
593 |
left_division: Left division index (1-20)
|
594 |
right_division: Right division index (1-20)
|
595 |
+
crop_types: List of crop types to include in visualization (e.g., ["16:9", "9:16"])
|
596 |
|
597 |
Returns:
|
598 |
+
PIL Image: Original image with selected crop regions visualized
|
599 |
"""
|
600 |
# Convert PIL Image to cv2 format
|
601 |
if isinstance(image, Image.Image):
|
|
|
618 |
cutout_image = image_cv[:, left_boundary:right_boundary].copy()
|
619 |
|
620 |
# Get YOLO detections for person bounding box
|
621 |
+
results = model(cutout_image, classes=[0], conf=0.6)
|
622 |
|
623 |
# Default values
|
624 |
cutout_center_x = cutout_image.shape[1] // 2
|
|
|
673 |
font_scale = 0.8
|
674 |
font_thickness = 2
|
675 |
|
676 |
+
# Draw cutout region (original divisions) if requested
|
677 |
+
if "cutout" in crop_types:
|
678 |
+
cv2.rectangle(
|
679 |
+
visualization,
|
680 |
+
(left_boundary, 0),
|
681 |
+
(right_boundary, height),
|
682 |
+
colors["cutout"],
|
683 |
+
thickness,
|
684 |
+
)
|
685 |
+
cv2.putText(
|
686 |
+
visualization,
|
687 |
+
"Cutout",
|
688 |
+
(left_boundary + 5, 30),
|
689 |
+
font,
|
690 |
+
font_scale,
|
691 |
+
colors["cutout"],
|
692 |
+
font_thickness,
|
693 |
+
)
|
694 |
|
695 |
+
# Create 16:9 version of the cutout if requested
|
696 |
+
if "16:9" in crop_types:
|
697 |
+
cutout_width = right_boundary - left_boundary
|
698 |
+
cutout_height = height
|
699 |
+
aspect_16_9 = 16 / 9
|
700 |
+
target_height_16_9 = int(cutout_width / aspect_16_9)
|
701 |
|
702 |
+
if target_height_16_9 <= height:
|
703 |
+
# Calculate 20% of person height for top margin
|
704 |
+
top_margin = int(original_person_height * 0.2)
|
|
|
|
|
|
|
705 |
|
706 |
+
# Start 20% above the person's top
|
707 |
+
y_start = int(max(0, original_person_top - top_margin))
|
708 |
|
709 |
+
# If this would make the crop exceed the bottom, adjust y_start
|
710 |
+
if y_start + target_height_16_9 > height:
|
711 |
+
y_start = int(max(0, height - target_height_16_9))
|
712 |
+
|
713 |
+
y_end = int(min(height, y_start + target_height_16_9))
|
714 |
+
|
715 |
+
cv2.rectangle(
|
716 |
+
visualization,
|
717 |
+
(left_boundary, y_start),
|
718 |
+
(right_boundary, y_end),
|
719 |
+
colors["16:9"],
|
720 |
+
thickness,
|
721 |
+
)
|
722 |
+
cv2.putText(
|
723 |
+
visualization,
|
724 |
+
"16:9",
|
725 |
+
(left_boundary + 5, y_start + 30),
|
726 |
+
font,
|
727 |
+
font_scale,
|
728 |
+
colors["16:9"],
|
729 |
+
font_thickness,
|
730 |
+
)
|
731 |
|
732 |
+
# Create 9:16 version if requested
|
733 |
+
if "9:16" in crop_types:
|
734 |
+
cutout_width = right_boundary - left_boundary
|
735 |
+
cutout_height = height
|
736 |
+
aspect_9_16 = 9 / 16
|
737 |
+
target_width_9_16 = int(cutout_height * aspect_9_16)
|
738 |
+
|
739 |
+
if target_width_9_16 <= cutout_width:
|
740 |
+
# Center horizontally around person
|
741 |
+
x_start = max(
|
742 |
+
0,
|
743 |
+
min(
|
744 |
+
left_boundary + cutout_width - target_width_9_16,
|
745 |
+
original_center_x - target_width_9_16 // 2,
|
746 |
+
),
|
747 |
+
)
|
748 |
+
x_end = x_start + target_width_9_16
|
749 |
+
cv2.rectangle(
|
750 |
+
visualization, (x_start, 0), (x_end, height), colors["9:16"], thickness
|
751 |
+
)
|
752 |
+
cv2.putText(
|
753 |
+
visualization,
|
754 |
+
"9:16",
|
755 |
+
(x_start + 5, 60),
|
756 |
+
font,
|
757 |
+
font_scale,
|
758 |
+
colors["9:16"],
|
759 |
+
font_thickness,
|
760 |
+
)
|
761 |
|
762 |
+
# Draw centered half width layout if requested
|
763 |
+
if "half" in crop_types:
|
764 |
+
half_width = width // 2
|
765 |
+
half_left_x = max(
|
766 |
+
0, min(width - half_width, original_center_x - half_width // 2)
|
767 |
+
)
|
768 |
+
half_right_x = half_left_x + half_width
|
769 |
cv2.rectangle(
|
770 |
visualization,
|
771 |
+
(half_left_x, 0),
|
772 |
+
(half_right_x, height),
|
773 |
+
colors["half"],
|
774 |
thickness,
|
775 |
)
|
776 |
cv2.putText(
|
777 |
visualization,
|
778 |
+
"Half Width",
|
779 |
+
(half_left_x + 5, 90),
|
780 |
font,
|
781 |
font_scale,
|
782 |
+
colors["half"],
|
783 |
font_thickness,
|
784 |
)
|
785 |
|
786 |
+
# Draw centered third width layout if requested
|
787 |
+
if "third" in crop_types:
|
788 |
+
third_width = width // 3
|
789 |
+
third_left_x = max(
|
790 |
+
0, min(width - third_width, original_center_x - third_width // 2)
|
|
|
|
|
|
|
|
|
|
|
|
|
791 |
)
|
792 |
+
third_right_x = third_left_x + third_width
|
793 |
cv2.rectangle(
|
794 |
+
visualization,
|
795 |
+
(third_left_x, 0),
|
796 |
+
(third_right_x, height),
|
797 |
+
colors["third"],
|
798 |
+
thickness,
|
799 |
)
|
800 |
cv2.putText(
|
801 |
visualization,
|
802 |
+
"Third Width",
|
803 |
+
(third_left_x + 5, 120),
|
804 |
font,
|
805 |
font_scale,
|
806 |
+
colors["third"],
|
807 |
font_thickness,
|
808 |
)
|
809 |
|
810 |
+
# Draw centered two-thirds width layout if requested
|
811 |
+
if "two_thirds" in crop_types:
|
812 |
+
two_thirds_width = (width * 2) // 3
|
813 |
+
two_thirds_left_x = max(
|
814 |
+
0, min(width - two_thirds_width, original_center_x - two_thirds_width // 2)
|
815 |
+
)
|
816 |
+
two_thirds_right_x = two_thirds_left_x + two_thirds_width
|
817 |
+
cv2.rectangle(
|
818 |
+
visualization,
|
819 |
+
(two_thirds_left_x, 0),
|
820 |
+
(two_thirds_right_x, height),
|
821 |
+
colors["two_thirds"],
|
822 |
+
thickness,
|
823 |
+
)
|
824 |
+
cv2.putText(
|
825 |
+
visualization,
|
826 |
+
"Two-Thirds Width",
|
827 |
+
(two_thirds_left_x + 5, 150),
|
828 |
+
font,
|
829 |
+
font_scale,
|
830 |
+
colors["two_thirds"],
|
831 |
+
font_thickness,
|
832 |
+
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
833 |
|
834 |
+
# Draw center point of person(s)
|
835 |
center_radius = 8
|
836 |
cv2.circle(
|
837 |
visualization,
|
838 |
+
(original_center_x, original_center_y),
|
839 |
center_radius,
|
840 |
(255, 255, 255),
|
841 |
-1,
|
842 |
)
|
843 |
cv2.circle(
|
844 |
+
visualization,
|
845 |
+
(original_center_x, original_center_y),
|
846 |
+
center_radius,
|
847 |
+
(0, 0, 0),
|
848 |
+
2,
|
849 |
)
|
850 |
cv2.putText(
|
851 |
visualization,
|
852 |
"Person Center",
|
853 |
+
(original_center_x + 10, original_center_y),
|
854 |
font,
|
855 |
font_scale,
|
856 |
(255, 255, 255),
|
|
|
870 |
Returns:
|
871 |
gr.Gallery: Gallery of all generated images
|
872 |
"""
|
873 |
+
# image_paths = get_sprite_firebase(cid, rsid, uid)
|
874 |
+
image_paths = ["data/sprite1.jpg", "data/sprite2.jpg"]
|
875 |
# Lists to store all images
|
876 |
all_images = []
|
877 |
all_captions = []
|
|
|
926 |
mid_image, left_division, right_division
|
927 |
)
|
928 |
|
929 |
+
# Create the first visualization with 16:9 and 9:16 crops only
|
930 |
+
aspect_ratio_visualization = draw_crops_on_original(
|
931 |
+
mid_image, left_division, right_division, ["16:9", "9:16"]
|
932 |
+
)
|
933 |
+
|
934 |
+
# Create the second visualization with the remaining layouts
|
935 |
+
other_layouts_visualization = draw_crops_on_original(
|
936 |
+
mid_image,
|
937 |
+
left_division,
|
938 |
+
right_division,
|
939 |
+
["cutout", "half", "third", "two_thirds"],
|
940 |
+
)
|
941 |
+
|
942 |
+
# Add visualizations showing crops
|
943 |
+
all_images.append(aspect_ratio_visualization)
|
944 |
+
all_captions.append(
|
945 |
+
f"Aspect Ratio Crops (16:9 & 9:16) {aspect_ratio_visualization.size}"
|
946 |
)
|
947 |
|
948 |
+
all_images.append(other_layouts_visualization)
|
949 |
+
all_captions.append(f"Other Layout Options {other_layouts_visualization.size}")
|
|
|
950 |
|
951 |
# Add input and middle image to gallery
|
952 |
all_images.append(input_image)
|