Spaces:
Runtime error
Runtime error
Add clip query
Browse files- ViT-B-32.pt +3 -0
- app.py +63 -14
ViT-B-32.pt
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:40d365715913c9da98579312b702a82c18be219cc2a73407c4526f58eba950af
|
3 |
+
size 353976522
|
app.py
CHANGED
@@ -1,8 +1,9 @@
|
|
1 |
import os
|
2 |
from functools import lru_cache
|
3 |
from random import randint
|
4 |
-
from typing import Dict, List
|
5 |
|
|
|
6 |
import cv2
|
7 |
import gradio as gr
|
8 |
import numpy as np
|
@@ -13,17 +14,27 @@ from segment_anything import SamAutomaticMaskGenerator, sam_model_registry
|
|
13 |
CHECKPOINT_PATH = "sam_vit_h_4b8939.pth"
|
14 |
MODEL_TYPE = "default"
|
15 |
MAX_WIDTH = MAX_HEIGHT = 800
|
|
|
16 |
THRESHOLD = 0.05
|
17 |
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
18 |
|
19 |
|
20 |
@lru_cache
|
21 |
-
def load_mask_generator(
|
22 |
sam = sam_model_registry[MODEL_TYPE](checkpoint=CHECKPOINT_PATH).to(device)
|
23 |
mask_generator = SamAutomaticMaskGenerator(sam)
|
24 |
return mask_generator
|
25 |
|
26 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
27 |
def adjust_image_size(image: np.ndarray) -> np.ndarray:
|
28 |
height, width = image.shape[:2]
|
29 |
if height > width:
|
@@ -36,23 +47,56 @@ def adjust_image_size(image: np.ndarray) -> np.ndarray:
|
|
36 |
return image
|
37 |
|
38 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
39 |
def filter_masks(
|
40 |
-
|
|
|
41 |
predicted_iou_threshold: float,
|
42 |
stability_score_threshold: float,
|
43 |
query: str,
|
44 |
clip_threshold: float,
|
45 |
-
) -> List[
|
46 |
-
|
|
|
|
|
47 |
for mask in masks:
|
48 |
if (
|
49 |
mask["predicted_iou"] < predicted_iou_threshold
|
50 |
or mask["stability_score"] < stability_score_threshold
|
51 |
):
|
52 |
continue
|
|
|
53 |
filtered_masks.append(mask)
|
54 |
|
55 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
56 |
|
57 |
|
58 |
def draw_masks(
|
@@ -62,7 +106,7 @@ def draw_masks(
|
|
62 |
color = [randint(127, 255) for _ in range(3)]
|
63 |
|
64 |
# draw mask overlay
|
65 |
-
colored_mask = np.expand_dims(mask, 0).repeat(3, axis=0)
|
66 |
colored_mask = np.moveaxis(colored_mask, 0, -1)
|
67 |
masked = np.ma.MaskedArray(image, mask=colored_mask, fill_value=color)
|
68 |
image_overlay = masked.filled()
|
@@ -70,7 +114,7 @@ def draw_masks(
|
|
70 |
|
71 |
# draw contour
|
72 |
contours, _ = cv2.findContours(
|
73 |
-
np.uint8(mask), cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE
|
74 |
)
|
75 |
cv2.drawContours(image, contours, -1, (255, 0, 0), 2)
|
76 |
return image
|
@@ -88,7 +132,12 @@ def segment(
|
|
88 |
image = adjust_image_size(cv2.imread(image_path))
|
89 |
masks = mask_generator.generate(image)
|
90 |
masks = filter_masks(
|
91 |
-
|
|
|
|
|
|
|
|
|
|
|
92 |
)
|
93 |
image = draw_masks(image, masks)
|
94 |
image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
|
@@ -112,16 +161,16 @@ demo = gr.Interface(
|
|
112 |
[
|
113 |
0.9,
|
114 |
0.8,
|
115 |
-
0.
|
116 |
os.path.join(os.path.dirname(__file__), "examples/dog.jpg"),
|
117 |
-
"",
|
118 |
],
|
119 |
[
|
120 |
0.9,
|
121 |
0.8,
|
122 |
-
0.
|
123 |
os.path.join(os.path.dirname(__file__), "examples/city.jpg"),
|
124 |
-
"",
|
125 |
],
|
126 |
[
|
127 |
0.9,
|
@@ -135,7 +184,7 @@ demo = gr.Interface(
|
|
135 |
0.8,
|
136 |
0.05,
|
137 |
os.path.join(os.path.dirname(__file__), "examples/horse.jpg"),
|
138 |
-
"",
|
139 |
],
|
140 |
],
|
141 |
)
|
|
|
1 |
import os
|
2 |
from functools import lru_cache
|
3 |
from random import randint
|
4 |
+
from typing import Any, Callable, Dict, List, Tuple
|
5 |
|
6 |
+
import clip
|
7 |
import cv2
|
8 |
import gradio as gr
|
9 |
import numpy as np
|
|
|
14 |
CHECKPOINT_PATH = "sam_vit_h_4b8939.pth"
|
15 |
MODEL_TYPE = "default"
|
16 |
MAX_WIDTH = MAX_HEIGHT = 800
|
17 |
+
CLIP_WIDTH = CLIP_HEIGHT = 300
|
18 |
THRESHOLD = 0.05
|
19 |
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
20 |
|
21 |
|
22 |
@lru_cache
|
23 |
+
def load_mask_generator() -> SamAutomaticMaskGenerator:
|
24 |
sam = sam_model_registry[MODEL_TYPE](checkpoint=CHECKPOINT_PATH).to(device)
|
25 |
mask_generator = SamAutomaticMaskGenerator(sam)
|
26 |
return mask_generator
|
27 |
|
28 |
|
29 |
+
@lru_cache
|
30 |
+
def load_clip(
|
31 |
+
name: str = "ViT-B-32.pt",
|
32 |
+
) -> Tuple[torch.nn.Module, Callable[[PIL.Image.Image], torch.Tensor]]:
|
33 |
+
model_path = os.path.join(".", name)
|
34 |
+
model, preprocess = clip.load(model_path, device=device)
|
35 |
+
return model.to(device), preprocess
|
36 |
+
|
37 |
+
|
38 |
def adjust_image_size(image: np.ndarray) -> np.ndarray:
|
39 |
height, width = image.shape[:2]
|
40 |
if height > width:
|
|
|
47 |
return image
|
48 |
|
49 |
|
50 |
+
@torch.no_grad()
|
51 |
+
def get_scores(crops: List[PIL.Image.Image], query: str) -> torch.Tensor:
|
52 |
+
model, preprocess = load_clip()
|
53 |
+
preprocessed = [preprocess(crop) for crop in crops]
|
54 |
+
preprocessed = torch.stack(preprocessed).to(device)
|
55 |
+
token = clip.tokenize(query).to(device)
|
56 |
+
img_features = model.encode_image(preprocessed)
|
57 |
+
txt_features = model.encode_text(token)
|
58 |
+
img_features /= img_features.norm(dim=-1, keepdim=True)
|
59 |
+
txt_features /= txt_features.norm(dim=-1, keepdim=True)
|
60 |
+
probs = 100.0 * img_features @ txt_features.T
|
61 |
+
return probs[:, 0].softmax(dim=0)
|
62 |
+
|
63 |
+
|
64 |
def filter_masks(
|
65 |
+
image: np.ndarray,
|
66 |
+
masks: List[Dict[str, Any]],
|
67 |
predicted_iou_threshold: float,
|
68 |
stability_score_threshold: float,
|
69 |
query: str,
|
70 |
clip_threshold: float,
|
71 |
+
) -> List[Dict[str, Any]]:
|
72 |
+
cropped_masks: List[PIL.Image.Image] = []
|
73 |
+
filtered_masks: List[Dict[str, Any]] = []
|
74 |
+
|
75 |
for mask in masks:
|
76 |
if (
|
77 |
mask["predicted_iou"] < predicted_iou_threshold
|
78 |
or mask["stability_score"] < stability_score_threshold
|
79 |
):
|
80 |
continue
|
81 |
+
|
82 |
filtered_masks.append(mask)
|
83 |
|
84 |
+
x, y, w, h = mask["bbox"]
|
85 |
+
crop = image[y : y + h, x : x + w]
|
86 |
+
crop = cv2.cvtColor(crop, cv2.COLOR_BGR2RGB)
|
87 |
+
crop = PIL.Image.fromarray(np.uint8(crop * 255)).convert("RGB")
|
88 |
+
crop.resize((CLIP_WIDTH, CLIP_HEIGHT))
|
89 |
+
cropped_masks.append(crop)
|
90 |
+
|
91 |
+
if query and filtered_masks:
|
92 |
+
scores = get_scores(cropped_masks, query)
|
93 |
+
filtered_masks = [
|
94 |
+
filtered_masks[i]
|
95 |
+
for i, score in enumerate(scores)
|
96 |
+
if score > clip_threshold
|
97 |
+
]
|
98 |
+
|
99 |
+
return filtered_masks
|
100 |
|
101 |
|
102 |
def draw_masks(
|
|
|
106 |
color = [randint(127, 255) for _ in range(3)]
|
107 |
|
108 |
# draw mask overlay
|
109 |
+
colored_mask = np.expand_dims(mask["segmentation"], 0).repeat(3, axis=0)
|
110 |
colored_mask = np.moveaxis(colored_mask, 0, -1)
|
111 |
masked = np.ma.MaskedArray(image, mask=colored_mask, fill_value=color)
|
112 |
image_overlay = masked.filled()
|
|
|
114 |
|
115 |
# draw contour
|
116 |
contours, _ = cv2.findContours(
|
117 |
+
np.uint8(mask["segmentation"]), cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE
|
118 |
)
|
119 |
cv2.drawContours(image, contours, -1, (255, 0, 0), 2)
|
120 |
return image
|
|
|
132 |
image = adjust_image_size(cv2.imread(image_path))
|
133 |
masks = mask_generator.generate(image)
|
134 |
masks = filter_masks(
|
135 |
+
image,
|
136 |
+
masks,
|
137 |
+
predicted_iou_threshold,
|
138 |
+
stability_score_threshold,
|
139 |
+
query,
|
140 |
+
clip_threshold,
|
141 |
)
|
142 |
image = draw_masks(image, masks)
|
143 |
image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
|
|
|
161 |
[
|
162 |
0.9,
|
163 |
0.8,
|
164 |
+
0.15,
|
165 |
os.path.join(os.path.dirname(__file__), "examples/dog.jpg"),
|
166 |
+
"A dog only",
|
167 |
],
|
168 |
[
|
169 |
0.9,
|
170 |
0.8,
|
171 |
+
0.1,
|
172 |
os.path.join(os.path.dirname(__file__), "examples/city.jpg"),
|
173 |
+
"A bridge on the water",
|
174 |
],
|
175 |
[
|
176 |
0.9,
|
|
|
184 |
0.8,
|
185 |
0.05,
|
186 |
os.path.join(os.path.dirname(__file__), "examples/horse.jpg"),
|
187 |
+
"horse",
|
188 |
],
|
189 |
],
|
190 |
)
|