Spaces:
Starting
on
T4
Starting
on
T4
AAAAAAyq
commited on
Commit
•
dc120b2
1
Parent(s):
77f39b8
Update app.py
Browse files- app.py +116 -119
- requirements.txt +1 -1
app.py
CHANGED
@@ -4,29 +4,70 @@ import matplotlib.pyplot as plt
|
|
4 |
import gradio as gr
|
5 |
import cv2
|
6 |
import torch
|
|
|
7 |
|
8 |
model = YOLO('checkpoints/FastSAM.pt') # load a custom model
|
9 |
|
10 |
|
11 |
-
def fast_process(annotations, image):
|
|
|
|
|
|
|
|
|
|
|
12 |
fig = plt.figure(figsize=(10, 10))
|
13 |
plt.imshow(image)
|
14 |
-
|
15 |
-
|
16 |
-
|
17 |
-
|
18 |
-
|
19 |
-
|
20 |
-
|
21 |
-
|
22 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
23 |
plt.axis('off')
|
24 |
plt.tight_layout()
|
25 |
return fig
|
26 |
|
27 |
|
28 |
# CPU post process
|
29 |
-
def fast_show_mask(annotation, ax
|
|
|
|
|
|
|
30 |
msak_sum = annotation.shape[0]
|
31 |
height = annotation.shape[1]
|
32 |
weight = annotation.shape[2]
|
@@ -36,136 +77,92 @@ def fast_show_mask(annotation, ax):
|
|
36 |
annotation = annotation[sorted_indices]
|
37 |
|
38 |
index = (annotation != 0).argmax(axis=0)
|
39 |
-
color = np.random.random((msak_sum,
|
40 |
-
transparency = np.ones((msak_sum,
|
41 |
-
visual = np.concatenate([color,
|
42 |
-
mask_image = np.expand_dims(annotation
|
43 |
|
44 |
-
show = np.zeros((height,
|
45 |
|
46 |
h_indices, w_indices = np.meshgrid(np.arange(height), np.arange(weight), indexing='ij')
|
47 |
indices = (index[h_indices, w_indices], h_indices, w_indices, slice(None))
|
48 |
# 使用向量化索引更新show的值
|
49 |
show[h_indices, w_indices, :] = mask_image[indices]
|
50 |
-
|
51 |
-
|
52 |
-
|
53 |
-
#
|
|
|
|
|
|
|
|
|
|
|
|
|
54 |
ax.imshow(show)
|
55 |
|
56 |
|
57 |
-
|
58 |
-
|
59 |
-
|
60 |
-
|
61 |
-
|
62 |
-
|
63 |
-
|
64 |
-
|
65 |
-
|
66 |
-
|
67 |
-
|
68 |
-
|
69 |
-
|
70 |
-
|
71 |
-
|
72 |
-
|
73 |
-
|
74 |
-
|
75 |
-
|
76 |
-
|
77 |
-
|
78 |
-
|
79 |
-
# ["assets/sa_561.jpg"], ["assets/sa_862.jpg"],
|
80 |
-
# ["assets/sa_1309.jpg"], ["assets/sa_8776.jpg"],
|
81 |
-
# ["assets/sa_10039.jpg"], ["assets/sa_11025.jpg"],],
|
82 |
-
)
|
83 |
-
|
84 |
-
demo.launch()
|
85 |
-
"""
|
86 |
-
|
87 |
-
from ultralytics import YOLO
|
88 |
-
import numpy as np
|
89 |
-
import matplotlib.pyplot as plt
|
90 |
-
import gradio as gr
|
91 |
-
import torch
|
92 |
-
|
93 |
-
model = YOLO('checkpoints/FastSAM.pt') # load a custom model
|
94 |
-
|
95 |
-
def format_results(result,filter = 0):
|
96 |
-
annotations = []
|
97 |
-
n = len(result.masks.data)
|
98 |
-
for i in range(n):
|
99 |
-
annotation = {}
|
100 |
-
mask = result.masks.data[i] == 1.0
|
101 |
-
|
102 |
-
if torch.sum(mask) < filter:
|
103 |
-
continue
|
104 |
-
annotation['id'] = i
|
105 |
-
annotation['segmentation'] = mask.cpu().numpy()
|
106 |
-
annotation['bbox'] = result.boxes.data[i]
|
107 |
-
annotation['score'] = result.boxes.conf[i]
|
108 |
-
annotation['area'] = annotation['segmentation'].sum()
|
109 |
-
annotations.append(annotation)
|
110 |
-
return annotations
|
111 |
-
|
112 |
-
def show_mask(annotation, ax, random_color=True, bbox=None, points=None):
|
113 |
-
if random_color : # random mask color
|
114 |
-
color = np.concatenate([np.random.random(3), np.array([0.6])], axis=0)
|
115 |
-
else:
|
116 |
-
color = np.array([30 / 255, 144 / 255, 255 / 255, 0.6])
|
117 |
-
if type(annotation) == dict:
|
118 |
-
annotation = annotation['segmentation']
|
119 |
-
mask = annotation
|
120 |
-
h, w = mask.shape[-2:]
|
121 |
-
mask_image = mask.reshape(h, w, 1) * color.reshape(1, 1, -1)
|
122 |
-
# draw box
|
123 |
if bbox is not None:
|
124 |
x1, y1, x2, y2 = bbox
|
125 |
ax.add_patch(plt.Rectangle((x1, y1), x2 - x1, y2 - y1, fill=False, edgecolor='b', linewidth=1))
|
126 |
# draw point
|
127 |
if points is not None:
|
128 |
-
|
129 |
-
|
130 |
-
|
131 |
-
|
132 |
-
def post_process(annotations, image, mask_random_color=True, bbox=None, points=None):
|
133 |
-
fig = plt.figure(figsize=(10, 10))
|
134 |
-
plt.imshow(image)
|
135 |
-
for i, mask in enumerate(annotations):
|
136 |
-
show_mask(mask, plt.gca(),random_color=mask_random_color,bbox=bbox,points=points)
|
137 |
-
plt.axis('off')
|
138 |
-
|
139 |
-
plt.tight_layout()
|
140 |
-
return fig
|
141 |
-
|
142 |
|
143 |
# post_process(results[0].masks, Image.open("../data/cake.png"))
|
144 |
|
145 |
-
def predict(input, input_size):
|
|
|
146 |
input_size = int(input_size) # 确保 imgsz 是整数
|
147 |
-
results = model(input, device=
|
148 |
-
|
149 |
-
|
150 |
-
pil_image = post_process(annotations=results, image=input)
|
151 |
return pil_image
|
152 |
|
|
|
|
|
153 |
# inp = 'assets/sa_192.jpg'
|
154 |
-
#
|
155 |
-
#
|
156 |
-
#
|
157 |
-
|
|
|
|
|
158 |
demo = gr.Interface(fn=predict,
|
159 |
-
inputs=[gr.
|
|
|
|
|
160 |
outputs=['plot'],
|
161 |
-
|
162 |
-
|
163 |
-
|
164 |
-
|
165 |
-
|
166 |
-
|
|
|
167 |
)
|
168 |
|
169 |
-
demo.launch()
|
170 |
-
|
171 |
-
"""
|
|
|
4 |
import gradio as gr
|
5 |
import cv2
|
6 |
import torch
|
7 |
+
from PIL import Image
|
8 |
|
9 |
model = YOLO('checkpoints/FastSAM.pt') # load a custom model
|
10 |
|
11 |
|
12 |
+
def fast_process(annotations, image, high_quality, device):
|
13 |
+
if isinstance(annotations[0],dict):
|
14 |
+
annotations = [annotation['segmentation'] for annotation in annotations]
|
15 |
+
|
16 |
+
original_h = image.height
|
17 |
+
original_w = image.width
|
18 |
fig = plt.figure(figsize=(10, 10))
|
19 |
plt.imshow(image)
|
20 |
+
if high_quality == True:
|
21 |
+
if isinstance(annotations[0],torch.Tensor):
|
22 |
+
annotations = np.array(annotations.cpu())
|
23 |
+
for i, mask in enumerate(annotations):
|
24 |
+
mask = cv2.morphologyEx(mask.astype(np.uint8), cv2.MORPH_CLOSE, np.ones((3, 3), np.uint8))
|
25 |
+
annotations[i] = cv2.morphologyEx(mask.astype(np.uint8), cv2.MORPH_OPEN, np.ones((8, 8), np.uint8))
|
26 |
+
if device == 'cpu':
|
27 |
+
annotations = np.array(annotations)
|
28 |
+
fast_show_mask(annotations,
|
29 |
+
plt.gca(),
|
30 |
+
bbox=None,
|
31 |
+
points=None,
|
32 |
+
pointlabel=None,
|
33 |
+
retinamask=True,
|
34 |
+
target_height=original_h,
|
35 |
+
target_width=original_w)
|
36 |
+
else:
|
37 |
+
if isinstance(annotations[0],np.ndarray):
|
38 |
+
annotations = torch.from_numpy(annotations)
|
39 |
+
fast_show_mask_gpu(annotations,
|
40 |
+
plt.gca(),
|
41 |
+
bbox=None,
|
42 |
+
points=None,
|
43 |
+
pointlabel=None)
|
44 |
+
if isinstance(annotations, torch.Tensor):
|
45 |
+
annotations = annotations.cpu().numpy()
|
46 |
+
if high_quality == True:
|
47 |
+
contour_all = []
|
48 |
+
temp = np.zeros((original_h, original_w,1))
|
49 |
+
for i, mask in enumerate(annotations):
|
50 |
+
if type(mask) == dict:
|
51 |
+
mask = mask['segmentation']
|
52 |
+
annotation = mask.astype(np.uint8)
|
53 |
+
contours, _ = cv2.findContours(annotation, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
|
54 |
+
for contour in contours:
|
55 |
+
contour_all.append(contour)
|
56 |
+
cv2.drawContours(temp, contour_all, -1, (255, 255, 255), 2)
|
57 |
+
color = np.array([0 / 255, 0 / 255, 255 / 255, 0.8])
|
58 |
+
contour_mask = temp / 225 * color.reshape(1, 1, -1)
|
59 |
+
plt.imshow(contour_mask)
|
60 |
+
|
61 |
plt.axis('off')
|
62 |
plt.tight_layout()
|
63 |
return fig
|
64 |
|
65 |
|
66 |
# CPU post process
|
67 |
+
def fast_show_mask(annotation, ax, bbox=None,
|
68 |
+
points=None, pointlabel=None,
|
69 |
+
retinamask=True, target_height=960,
|
70 |
+
target_width=960):
|
71 |
msak_sum = annotation.shape[0]
|
72 |
height = annotation.shape[1]
|
73 |
weight = annotation.shape[2]
|
|
|
77 |
annotation = annotation[sorted_indices]
|
78 |
|
79 |
index = (annotation != 0).argmax(axis=0)
|
80 |
+
color = np.random.random((msak_sum,1,1,3))
|
81 |
+
transparency = np.ones((msak_sum,1,1,1)) * 0.6
|
82 |
+
visual = np.concatenate([color,transparency],axis=-1)
|
83 |
+
mask_image = np.expand_dims(annotation,-1) * visual
|
84 |
|
85 |
+
show = np.zeros((height,weight,4))
|
86 |
|
87 |
h_indices, w_indices = np.meshgrid(np.arange(height), np.arange(weight), indexing='ij')
|
88 |
indices = (index[h_indices, w_indices], h_indices, w_indices, slice(None))
|
89 |
# 使用向量化索引更新show的值
|
90 |
show[h_indices, w_indices, :] = mask_image[indices]
|
91 |
+
if bbox is not None:
|
92 |
+
x1, y1, x2, y2 = bbox
|
93 |
+
ax.add_patch(plt.Rectangle((x1, y1), x2 - x1, y2 - y1, fill=False, edgecolor='b', linewidth=1))
|
94 |
+
# draw point
|
95 |
+
if points is not None:
|
96 |
+
plt.scatter([point[0] for i, point in enumerate(points) if pointlabel[i]==1], [point[1] for i, point in enumerate(points) if pointlabel[i]==1], s=20, c='y')
|
97 |
+
plt.scatter([point[0] for i, point in enumerate(points) if pointlabel[i]==0], [point[1] for i, point in enumerate(points) if pointlabel[i]==0], s=20, c='m')
|
98 |
+
|
99 |
+
if retinamask==False:
|
100 |
+
show = cv2.resize(show,(target_width,target_height),interpolation=cv2.INTER_NEAREST)
|
101 |
ax.imshow(show)
|
102 |
|
103 |
|
104 |
+
def fast_show_mask_gpu(annotation, ax,
|
105 |
+
bbox=None, points=None,
|
106 |
+
pointlabel=None):
|
107 |
+
msak_sum = annotation.shape[0]
|
108 |
+
height = annotation.shape[1]
|
109 |
+
weight = annotation.shape[2]
|
110 |
+
areas = torch.sum(annotation, dim=(1, 2))
|
111 |
+
sorted_indices = torch.argsort(areas, descending=False)
|
112 |
+
annotation = annotation[sorted_indices]
|
113 |
+
# 找每个位置第一个非零值下标
|
114 |
+
index = (annotation != 0).to(torch.long).argmax(dim=0)
|
115 |
+
color = torch.rand((msak_sum,1,1,3)).to(annotation.device)
|
116 |
+
transparency = torch.ones((msak_sum,1,1,1)).to(annotation.device) * 0.6
|
117 |
+
visual = torch.cat([color,transparency],dim=-1)
|
118 |
+
mask_image = torch.unsqueeze(annotation,-1) * visual
|
119 |
+
# 按index取数,index指每个位置选哪个batch的数,把mask_image转成一个batch的形式
|
120 |
+
show = torch.zeros((height,weight,4)).to(annotation.device)
|
121 |
+
h_indices, w_indices = torch.meshgrid(torch.arange(height), torch.arange(weight))
|
122 |
+
indices = (index[h_indices, w_indices], h_indices, w_indices, slice(None))
|
123 |
+
# 使用向量化索引更新show的值
|
124 |
+
show[h_indices, w_indices, :] = mask_image[indices]
|
125 |
+
show_cpu = show.cpu().numpy()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
126 |
if bbox is not None:
|
127 |
x1, y1, x2, y2 = bbox
|
128 |
ax.add_patch(plt.Rectangle((x1, y1), x2 - x1, y2 - y1, fill=False, edgecolor='b', linewidth=1))
|
129 |
# draw point
|
130 |
if points is not None:
|
131 |
+
plt.scatter([point[0] for i, point in enumerate(points) if pointlabel[i]==1], [point[1] for i, point in enumerate(points) if pointlabel[i]==1], s=20, c='y')
|
132 |
+
plt.scatter([point[0] for i, point in enumerate(points) if pointlabel[i]==0], [point[1] for i, point in enumerate(points) if pointlabel[i]==0], s=20, c='m')
|
133 |
+
ax.imshow(show_cpu)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
134 |
|
135 |
# post_process(results[0].masks, Image.open("../data/cake.png"))
|
136 |
|
137 |
+
def predict(input, input_size=512, high_quality_visual=True):
|
138 |
+
device = 'cuda' if torch.cuda.is_available() else 'cpu'
|
139 |
input_size = int(input_size) # 确保 imgsz 是整数
|
140 |
+
results = model(input, device=device, retina_masks=True, iou=0.7, conf=0.25, imgsz=input_size)
|
141 |
+
pil_image = fast_process(annotations=results[0].masks.data,
|
142 |
+
image=input, high_quality=high_quality_visual, device=device)
|
|
|
143 |
return pil_image
|
144 |
|
145 |
+
# input_size=1024
|
146 |
+
# high_quality_visual=True
|
147 |
# inp = 'assets/sa_192.jpg'
|
148 |
+
# input = Image.open(inp)
|
149 |
+
# device = 'cuda' if torch.cuda.is_available() else 'cpu'
|
150 |
+
# input_size = int(input_size) # 确保 imgsz 是整数
|
151 |
+
# results = model(input, device=device, retina_masks=True, iou=0.7, conf=0.25, imgsz=input_size)
|
152 |
+
# pil_image = fast_process(annotations=results[0].masks.data,
|
153 |
+
# image=input, high_quality=high_quality_visual, device=device)
|
154 |
demo = gr.Interface(fn=predict,
|
155 |
+
inputs=[gr.components.Image(type='pil'),
|
156 |
+
gr.components.Dropdown(choices=[512, 800, 1024], default=512),
|
157 |
+
gr.components.Checkbox(default=True)],
|
158 |
outputs=['plot'],
|
159 |
+
# examples=[["assets/sa_8776.jpg", 1024, True]],
|
160 |
+
# ["assets/sa_1309.jpg", 1024]],
|
161 |
+
examples=[["assets/sa_192.jpg", 1024, True], ["assets/sa_414.jpg", 1024, True],
|
162 |
+
["assets/sa_561.jpg", 1024, True], ["assets/sa_862.jpg", 1024, True],
|
163 |
+
["assets/sa_1309.jpg", 1024, True], ["assets/sa_8776.jpg", 1024, True],
|
164 |
+
["assets/sa_10039.jpg", 1024, True], ["assets/sa_11025.jpg", 1024, True],],
|
165 |
+
cache_examples=False,
|
166 |
)
|
167 |
|
168 |
+
demo.launch()
|
|
|
|
requirements.txt
CHANGED
@@ -1,7 +1,7 @@
|
|
1 |
# Base-----------------------------------
|
2 |
matplotlib==3.2.2
|
3 |
numpy
|
4 |
-
|
5 |
# Pillow>=7.1.2
|
6 |
# PyYAML>=5.3.1
|
7 |
# requests>=2.23.0
|
|
|
1 |
# Base-----------------------------------
|
2 |
matplotlib==3.2.2
|
3 |
numpy
|
4 |
+
opencv-python
|
5 |
# Pillow>=7.1.2
|
6 |
# PyYAML>=5.3.1
|
7 |
# requests>=2.23.0
|