chore: upgrade moel
Browse files- model.onnx +1 -1
- src/sscd.py +0 -42
model.onnx
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
size 235
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:abe326cb527103d52edd6d5fba82f8dfa34d1012d1f005b298c652434d283d39
|
3 |
size 235
|
src/sscd.py
DELETED
@@ -1,42 +0,0 @@
|
|
1 |
-
from torchvision import transforms
|
2 |
-
import torch
|
3 |
-
from PIL import Image
|
4 |
-
import torch.nn.functional as F
|
5 |
-
from matplotlib import pyplot as plt
|
6 |
-
|
7 |
-
##### Global variable
|
8 |
-
normalize = transforms.Normalize(
|
9 |
-
mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225],
|
10 |
-
)
|
11 |
-
preproccess = transforms.Compose([
|
12 |
-
transforms.Resize(288),
|
13 |
-
transforms.ToTensor(),
|
14 |
-
normalize,
|
15 |
-
])
|
16 |
-
|
17 |
-
model = torch.jit.load("./msscddiscmixup.torchscript.pt")
|
18 |
-
|
19 |
-
|
20 |
-
def visualize(path):
|
21 |
-
image = Image.open(path)
|
22 |
-
plt.figure(figsize=(10, 10))
|
23 |
-
plt.axis('off')
|
24 |
-
plt.imshow(image)
|
25 |
-
|
26 |
-
|
27 |
-
def extract_feature(img_path):
|
28 |
-
img = Image.open(img_path).convert('RGB')
|
29 |
-
batch = preproccess(img).unsqueeze(0)
|
30 |
-
return model(batch)[0, :]
|
31 |
-
|
32 |
-
|
33 |
-
def simi(img1, img2):
|
34 |
-
visualize(img1)
|
35 |
-
visualize(img2)
|
36 |
-
vec1 = extract_feature(img1)
|
37 |
-
vec2 = extract_feature(img2)
|
38 |
-
# subtract the mean and then L2 norm
|
39 |
-
cos_sim = F.cosine_similarity(vec1, vec2, dim=0)
|
40 |
-
# 余弦相似度得分大于0.75,匹配准确度90%
|
41 |
-
print('similarity:', cos_sim)
|
42 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|