Update app.py
Browse files
app.py
CHANGED
@@ -1,91 +1,123 @@
|
|
1 |
-
from segment_anything import sam_model_registry, SamPredictor
|
2 |
-
import torch
|
3 |
import cv2
|
4 |
import numpy as np
|
5 |
-
import gradio as gr
|
6 |
import pandas as pd
|
|
|
|
|
7 |
import matplotlib.pyplot as plt
|
|
|
|
|
|
|
|
|
|
|
8 |
|
9 |
-
#
|
10 |
-
|
11 |
-
|
12 |
-
|
13 |
-
|
14 |
-
|
|
|
|
|
|
|
15 |
|
16 |
def preprocess_image(image):
|
17 |
-
"""Convert image to
|
18 |
if len(image.shape) == 2:
|
19 |
image = cv2.cvtColor(image, cv2.COLOR_GRAY2RGB)
|
20 |
-
return image
|
21 |
|
22 |
-
|
23 |
-
|
24 |
-
|
25 |
-
|
26 |
-
|
27 |
-
# Generate automatic masks (SAM can also take prompts for guided segmentation)
|
28 |
-
masks, _, _ = predictor.predict(
|
29 |
-
point_coords=None,
|
30 |
-
point_labels=None,
|
31 |
-
multimask_output=True
|
32 |
)
|
33 |
|
34 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
35 |
features = []
|
|
|
|
|
36 |
for i, mask in enumerate(masks):
|
37 |
-
|
38 |
-
contours, _ = cv2.findContours(
|
39 |
|
40 |
-
for
|
41 |
area = cv2.contourArea(contour)
|
42 |
perimeter = cv2.arcLength(contour, True)
|
43 |
circularity = 4 * np.pi * area / (perimeter * perimeter) if perimeter > 0 else 0
|
44 |
|
|
|
45 |
if 100 < area < 5000 and circularity > 0.7:
|
46 |
M = cv2.moments(contour)
|
47 |
if M["m00"] != 0:
|
48 |
cx = int(M["m10"] / M["m00"])
|
49 |
cy = int(M["m01"] / M["m00"])
|
50 |
-
features.append(
|
51 |
-
|
52 |
-
|
53 |
-
|
54 |
-
|
55 |
-
|
56 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
57 |
|
58 |
def process_image(image):
|
59 |
if image is None:
|
60 |
return None, None, None, None
|
61 |
|
62 |
-
|
63 |
-
vis_img = image.copy()
|
64 |
-
|
65 |
-
for feature in features:
|
66 |
-
contour = contours[int(feature['label'].split('-')[1]) - 1]
|
67 |
-
cv2.drawContours(vis_img, [contour], -1, (0, 255, 0), 2)
|
68 |
-
cv2.putText(vis_img, str(feature['label']), (feature['centroid_x'], feature['centroid_y']),
|
69 |
-
cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 0, 255), 1)
|
70 |
-
|
71 |
df = pd.DataFrame(features)
|
72 |
-
|
|
|
73 |
|
74 |
def analyze(image):
|
75 |
-
|
76 |
|
77 |
-
plt.style.use(
|
78 |
fig, axes = plt.subplots(1, 2, figsize=(12, 5))
|
79 |
|
80 |
if not df.empty:
|
81 |
-
axes[0].hist(df[
|
82 |
-
axes[0].set_title(
|
83 |
|
84 |
-
axes[1].scatter(df[
|
85 |
-
axes[1].set_title(
|
86 |
|
87 |
-
return
|
88 |
|
89 |
# Gradio Interface
|
90 |
-
demo = gr.Interface(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
91 |
demo.launch()
|
|
|
|
|
|
|
1 |
import cv2
|
2 |
import numpy as np
|
|
|
3 |
import pandas as pd
|
4 |
+
import torch
|
5 |
+
import gradio as gr
|
6 |
import matplotlib.pyplot as plt
|
7 |
+
from segment_anything import sam_model_registry, SamAutomaticMaskGenerator
|
8 |
+
|
9 |
+
# Ensure the SAM model checkpoint is downloaded
|
10 |
+
SAM_CHECKPOINT = "sam_vit_h.pth" # Update path if needed
|
11 |
+
MODEL_TYPE = "vit_h"
|
12 |
|
13 |
+
# Check if CUDA is available for GPU processing
|
14 |
+
DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
|
15 |
+
|
16 |
+
# Load the SAM model
|
17 |
+
try:
|
18 |
+
sam = sam_model_registry[MODEL_TYPE](checkpoint=SAM_CHECKPOINT).to(DEVICE)
|
19 |
+
mask_generator = SamAutomaticMaskGenerator(sam)
|
20 |
+
except FileNotFoundError:
|
21 |
+
raise FileNotFoundError(f"Checkpoint file '{SAM_CHECKPOINT}' not found. Download it from: https://github.com/facebookresearch/segment-anything")
|
22 |
|
23 |
def preprocess_image(image):
|
24 |
+
"""Convert image to grayscale and apply adaptive thresholding for better cell detection."""
|
25 |
if len(image.shape) == 2:
|
26 |
image = cv2.cvtColor(image, cv2.COLOR_GRAY2RGB)
|
|
|
27 |
|
28 |
+
gray = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)
|
29 |
+
|
30 |
+
# Apply adaptive thresholding for better contrast
|
31 |
+
adaptive_thresh = cv2.adaptiveThreshold(
|
32 |
+
gray, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, cv2.THRESH_BINARY_INV, 11, 2
|
|
|
|
|
|
|
|
|
|
|
33 |
)
|
34 |
|
35 |
+
# Morphological operations to remove noise
|
36 |
+
kernel = np.ones((3, 3), np.uint8)
|
37 |
+
clean_mask = cv2.morphologyEx(adaptive_thresh, cv2.MORPH_CLOSE, kernel, iterations=2)
|
38 |
+
clean_mask = cv2.morphologyEx(clean_mask, cv2.MORPH_OPEN, kernel, iterations=2)
|
39 |
+
|
40 |
+
return clean_mask
|
41 |
+
|
42 |
+
def detect_blood_cells(image):
|
43 |
+
"""Detect blood cells using SAM segmentation + contour analysis."""
|
44 |
+
# Generate masks using SAM
|
45 |
+
masks = mask_generator.generate(image)
|
46 |
+
|
47 |
features = []
|
48 |
+
processed_image = image.copy()
|
49 |
+
|
50 |
for i, mask in enumerate(masks):
|
51 |
+
mask_binary = mask["segmentation"].astype(np.uint8) * 255 # Convert to binary
|
52 |
+
contours, _ = cv2.findContours(mask_binary, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
|
53 |
|
54 |
+
for contour in contours:
|
55 |
area = cv2.contourArea(contour)
|
56 |
perimeter = cv2.arcLength(contour, True)
|
57 |
circularity = 4 * np.pi * area / (perimeter * perimeter) if perimeter > 0 else 0
|
58 |
|
59 |
+
# Filter small or irregular shapes
|
60 |
if 100 < area < 5000 and circularity > 0.7:
|
61 |
M = cv2.moments(contour)
|
62 |
if M["m00"] != 0:
|
63 |
cx = int(M["m10"] / M["m00"])
|
64 |
cy = int(M["m01"] / M["m00"])
|
65 |
+
features.append(
|
66 |
+
{
|
67 |
+
"label": len(features) + 1,
|
68 |
+
"area": area,
|
69 |
+
"perimeter": perimeter,
|
70 |
+
"circularity": circularity,
|
71 |
+
"centroid_x": cx,
|
72 |
+
"centroid_y": cy,
|
73 |
+
}
|
74 |
+
)
|
75 |
+
|
76 |
+
# Draw detected cell on image
|
77 |
+
cv2.drawContours(processed_image, [contour], -1, (0, 255, 0), 2)
|
78 |
+
cv2.putText(
|
79 |
+
processed_image,
|
80 |
+
str(len(features)),
|
81 |
+
(cx, cy),
|
82 |
+
cv2.FONT_HERSHEY_SIMPLEX,
|
83 |
+
0.5,
|
84 |
+
(0, 0, 255),
|
85 |
+
1,
|
86 |
+
)
|
87 |
+
|
88 |
+
return processed_image, features
|
89 |
|
90 |
def process_image(image):
|
91 |
if image is None:
|
92 |
return None, None, None, None
|
93 |
|
94 |
+
processed_img, features = detect_blood_cells(image)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
95 |
df = pd.DataFrame(features)
|
96 |
+
|
97 |
+
return processed_img, df
|
98 |
|
99 |
def analyze(image):
|
100 |
+
processed_img, df = process_image(image)
|
101 |
|
102 |
+
plt.style.use("dark_background")
|
103 |
fig, axes = plt.subplots(1, 2, figsize=(12, 5))
|
104 |
|
105 |
if not df.empty:
|
106 |
+
axes[0].hist(df["area"], bins=20, color="cyan", edgecolor="black")
|
107 |
+
axes[0].set_title("Cell Size Distribution")
|
108 |
|
109 |
+
axes[1].scatter(df["area"], df["circularity"], alpha=0.6, c="magenta")
|
110 |
+
axes[1].set_title("Area vs Circularity")
|
111 |
|
112 |
+
return processed_img, fig, df
|
113 |
|
114 |
# Gradio Interface
|
115 |
+
demo = gr.Interface(
|
116 |
+
fn=analyze,
|
117 |
+
inputs=gr.Image(type="numpy"),
|
118 |
+
outputs=[gr.Image(), gr.Plot(), gr.Dataframe()],
|
119 |
+
title="Blood Cell Detection",
|
120 |
+
description="Detect and analyze blood cells using SAM segmentation & contour analysis.",
|
121 |
+
)
|
122 |
+
|
123 |
demo.launch()
|