SauravMaheshkar
commited on
Commit
•
6038d30
1
Parent(s):
8a9151b
feat: default to mask generation when no annotations are provided
Browse files- app.py +28 -21
- src/plot_utils.py +68 -47
app.py
CHANGED
@@ -5,6 +5,7 @@ import gradio as gr
|
|
5 |
import numpy as np
|
6 |
from gradio_image_annotation import image_annotator
|
7 |
from sam2 import load_model
|
|
|
8 |
from sam2.sam2_image_predictor import SAM2ImagePredictor
|
9 |
|
10 |
from src.plot_utils import export_mask
|
@@ -17,30 +18,36 @@ def predict(model_choice, annotations: Dict[str, Any]):
|
|
17 |
ckpt_path=f"assets/checkpoints/sam2_hiera_{model_choice}.pt",
|
18 |
device="cpu",
|
19 |
)
|
20 |
-
|
21 |
-
|
22 |
-
|
23 |
-
|
24 |
-
|
25 |
-
|
26 |
-
|
27 |
-
|
28 |
-
|
29 |
-
|
30 |
-
|
|
|
31 |
|
32 |
-
|
33 |
-
|
34 |
-
|
35 |
-
|
36 |
-
|
37 |
-
|
|
|
|
|
|
|
|
|
38 |
|
39 |
-
|
40 |
-
# handle single mask cases
|
41 |
-
masks = np.expand_dims(masks, axis=0)
|
42 |
|
43 |
-
|
|
|
|
|
|
|
44 |
|
45 |
|
46 |
with gr.Blocks(delete_cache=(30, 30)) as demo:
|
|
|
5 |
import numpy as np
|
6 |
from gradio_image_annotation import image_annotator
|
7 |
from sam2 import load_model
|
8 |
+
from sam2.automatic_mask_generator import SAM2AutomaticMaskGenerator
|
9 |
from sam2.sam2_image_predictor import SAM2ImagePredictor
|
10 |
|
11 |
from src.plot_utils import export_mask
|
|
|
18 |
ckpt_path=f"assets/checkpoints/sam2_hiera_{model_choice}.pt",
|
19 |
device="cpu",
|
20 |
)
|
21 |
+
if annotations["boxes"]:
|
22 |
+
predictor = SAM2ImagePredictor(sam2_model) # type:ignore
|
23 |
+
predictor.set_image(annotations["image"])
|
24 |
+
coordinates = []
|
25 |
+
for i in range(len(annotations["boxes"])):
|
26 |
+
coordinate = [
|
27 |
+
int(annotations["boxes"][i]["xmin"]),
|
28 |
+
int(annotations["boxes"][i]["ymin"]),
|
29 |
+
int(annotations["boxes"][i]["xmax"]),
|
30 |
+
int(annotations["boxes"][i]["ymax"]),
|
31 |
+
]
|
32 |
+
coordinates.append(coordinate)
|
33 |
|
34 |
+
masks, scores, _ = predictor.predict(
|
35 |
+
point_coords=None,
|
36 |
+
point_labels=None,
|
37 |
+
box=np.array(coordinates),
|
38 |
+
multimask_output=False,
|
39 |
+
)
|
40 |
+
|
41 |
+
if masks.shape[0] == 1:
|
42 |
+
# handle single mask cases
|
43 |
+
masks = np.expand_dims(masks, axis=0)
|
44 |
|
45 |
+
return export_mask(masks)
|
|
|
|
|
46 |
|
47 |
+
else:
|
48 |
+
mask_generator = SAM2AutomaticMaskGenerator(sam2_model) # type: ignore
|
49 |
+
masks = mask_generator.generate(annotations["image"])
|
50 |
+
return export_mask(masks, autogenerated=True)
|
51 |
|
52 |
|
53 |
with gr.Blocks(delete_cache=(30, 30)) as demo:
|
src/plot_utils.py
CHANGED
@@ -1,68 +1,89 @@
|
|
1 |
from typing import Optional
|
2 |
|
|
|
3 |
import numpy as np
|
4 |
from PIL import Image
|
5 |
|
6 |
|
7 |
def export_mask(
|
8 |
-
masks
|
|
|
9 |
random_color: Optional[bool] = True,
|
10 |
smoothen_contours: Optional[bool] = True,
|
11 |
) -> Image:
|
12 |
-
|
13 |
-
|
|
|
14 |
|
15 |
-
|
16 |
-
|
17 |
|
18 |
-
|
19 |
-
|
20 |
|
21 |
-
|
22 |
-
|
23 |
-
|
24 |
-
|
25 |
|
26 |
-
|
27 |
-
|
28 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
29 |
else:
|
30 |
-
|
31 |
-
|
32 |
-
|
|
|
33 |
|
34 |
-
|
35 |
-
|
|
|
|
|
36 |
|
37 |
-
|
38 |
-
|
39 |
-
color_image[combined_mask == i] = mask_color
|
40 |
|
41 |
-
|
42 |
-
|
43 |
|
44 |
-
|
45 |
-
if smoothen_contours:
|
46 |
-
import cv2
|
47 |
|
48 |
-
contours_image = np.zeros((h, w, 4), dtype=np.float32)
|
49 |
|
50 |
-
|
51 |
-
|
52 |
-
|
53 |
-
|
54 |
-
|
55 |
-
|
56 |
-
|
57 |
-
|
58 |
-
|
59 |
-
contours_image = cv2.drawContours(
|
60 |
-
contours_image, contours, -1, (0, 0, 0, 0.5), thickness=2
|
61 |
-
)
|
62 |
-
|
63 |
-
# Convert contours to PIL image and blend with the color image
|
64 |
-
contours_image = (contours_image[:, :, :3] * 255).astype(np.uint8)
|
65 |
-
contours_pil_image = Image.fromarray(contours_image)
|
66 |
-
pil_image = Image.blend(pil_image, contours_pil_image, alpha=0.6)
|
67 |
-
|
68 |
-
return pil_image
|
|
|
1 |
from typing import Optional
|
2 |
|
3 |
+
import cv2
|
4 |
import numpy as np
|
5 |
from PIL import Image
|
6 |
|
7 |
|
8 |
def export_mask(
|
9 |
+
masks,
|
10 |
+
autogenerated: Optional[bool] = False,
|
11 |
random_color: Optional[bool] = True,
|
12 |
smoothen_contours: Optional[bool] = True,
|
13 |
) -> Image:
|
14 |
+
if not autogenerated:
|
15 |
+
num_masks, _, h, w = masks.shape
|
16 |
+
num_masks = len(masks)
|
17 |
|
18 |
+
# Ensure masks are 2D by squeezing channel dimension
|
19 |
+
masks = masks.squeeze(axis=1)
|
20 |
|
21 |
+
# Create a single uint8 image with unique values for each mask
|
22 |
+
combined_mask = np.zeros((h, w), dtype=np.uint8)
|
23 |
|
24 |
+
for i in range(num_masks):
|
25 |
+
mask = masks[i]
|
26 |
+
mask = mask.astype(np.uint8)
|
27 |
+
combined_mask[mask > 0] = i + 1
|
28 |
|
29 |
+
# Create color map for visualization
|
30 |
+
if random_color:
|
31 |
+
colors = np.random.rand(num_masks, 3) # Random colors for each mask
|
32 |
+
else:
|
33 |
+
colors = np.array(
|
34 |
+
[[30 / 255, 144 / 255, 255 / 255]] * num_masks
|
35 |
+
) # Use fixed color
|
36 |
+
|
37 |
+
# Create an RGB image where each mask has its own color
|
38 |
+
color_image = np.zeros((h, w, 3), dtype=np.uint8)
|
39 |
+
|
40 |
+
for i in range(1, num_masks + 1):
|
41 |
+
mask_color = colors[i - 1] * 255
|
42 |
+
color_image[combined_mask == i] = mask_color
|
43 |
+
|
44 |
+
# Convert the NumPy array to a PIL Image
|
45 |
+
pil_image = Image.fromarray(color_image)
|
46 |
+
|
47 |
+
# Optional: Add contours to the mask image
|
48 |
+
if smoothen_contours:
|
49 |
+
contours_image = np.zeros((h, w, 4), dtype=np.float32)
|
50 |
+
|
51 |
+
for i in range(1, num_masks + 1):
|
52 |
+
mask = (combined_mask == i).astype(np.uint8)
|
53 |
+
contours_image = smoothen(mask, contours_image)
|
54 |
+
|
55 |
+
# Convert contours to PIL image and blend with the color image
|
56 |
+
contours_image = (contours_image[:, :, :3] * 255).astype(np.uint8)
|
57 |
+
contours_pil_image = Image.fromarray(contours_image)
|
58 |
+
pil_image = Image.blend(pil_image, contours_pil_image, alpha=0.6)
|
59 |
+
|
60 |
+
return pil_image
|
61 |
else:
|
62 |
+
sorted_anns = sorted(masks, key=(lambda x: x["area"]), reverse=True)
|
63 |
+
img_shape = sorted_anns[0]["segmentation"].shape
|
64 |
+
img = np.ones((img_shape[0], img_shape[1], 4))
|
65 |
+
img[:, :, 3] = 0
|
66 |
|
67 |
+
for ann in sorted_anns:
|
68 |
+
m = ann["segmentation"]
|
69 |
+
color_mask = np.concatenate([np.random.random(3), [0.5]])
|
70 |
+
img[m] = color_mask
|
71 |
|
72 |
+
if smoothen_contours:
|
73 |
+
img = smoothen(m, img)
|
|
|
74 |
|
75 |
+
img = (img * 255).astype(np.uint8)
|
76 |
+
pil_image = Image.fromarray(img)
|
77 |
|
78 |
+
return pil_image
|
|
|
|
|
79 |
|
|
|
80 |
|
81 |
+
def smoothen(mask: np.ndarray, image: np.ndarray) -> np.ndarray:
|
82 |
+
contours, _ = cv2.findContours(
|
83 |
+
mask.astype(np.uint8), cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE
|
84 |
+
)
|
85 |
+
contours = [
|
86 |
+
cv2.approxPolyDP(contour, epsilon=0.01, closed=True) for contour in contours
|
87 |
+
]
|
88 |
+
image = cv2.drawContours(image, contours, -1, (0, 0, 1, 0.4), thickness=1)
|
89 |
+
return image
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|