Commit
•
ecc6de5
1
Parent(s):
d29ec33
Update image_validator.py
Browse files- image_validator.py +62 -63
image_validator.py
CHANGED
@@ -1,64 +1,63 @@
|
|
1 |
-
from transformers import CLIPProcessor, CLIPModel, ViTImageProcessor, ViTModel
|
2 |
-
from PIL import Image
|
3 |
-
from sklearn.metrics.pairwise import cosine_similarity
|
4 |
-
|
5 |
-
from warnings import filterwarnings
|
6 |
-
filterwarnings("ignore")
|
7 |
-
|
8 |
-
models = ["CLIP-ViT Base", "ViT Base", "DINO ViT-S16"]
|
9 |
-
models_info = {
|
10 |
-
"CLIP-ViT Base": {
|
11 |
-
"model_size": "386MB",
|
12 |
-
"model_url": "openai/clip-vit-base-patch32",
|
13 |
-
"efficiency": "High",
|
14 |
-
},
|
15 |
-
"ViT Base": {
|
16 |
-
"model_size": "304MB",
|
17 |
-
"model_url": "google/vit-base-patch16-224",
|
18 |
-
"efficiency": "High",
|
19 |
-
},
|
20 |
-
"DINO ViT-S16": {
|
21 |
-
"model_size": "1.34GB",
|
22 |
-
"model_url": "facebook/dino-vits16",
|
23 |
-
"efficiency": "Moderate",
|
24 |
-
},
|
25 |
-
}
|
26 |
-
|
27 |
-
class Image_Validator:
|
28 |
-
def __init__(self, model_name=None):
|
29 |
-
if model_name is None: model_name="ViT Base"
|
30 |
-
|
31 |
-
self.model_info = models_info[model_name]
|
32 |
-
model_url = self.model_info["model_url"]
|
33 |
-
|
34 |
-
if model_name == "CLIP-ViT Base":
|
35 |
-
self.model = CLIPModel.from_pretrained(model_url)
|
36 |
-
self.processor = CLIPProcessor.from_pretrained(model_url)
|
37 |
-
|
38 |
-
elif model_name == "ViT Base":
|
39 |
-
self.model = ViTModel.from_pretrained(model_url)
|
40 |
-
self.feature_extractor = ViTImageProcessor.from_pretrained(model_url)
|
41 |
-
|
42 |
-
elif model_name == "DINO ViT-S16":
|
43 |
-
self.model = ViTModel.from_pretrained(model_url)
|
44 |
-
self.feature_extractor = ViTImageProcessor.from_pretrained(model_url)
|
45 |
-
|
46 |
-
def get_image_embedding(self,
|
47 |
-
|
48 |
-
|
49 |
-
|
50 |
-
|
51 |
-
|
52 |
-
|
53 |
-
|
54 |
-
|
55 |
-
|
56 |
-
|
57 |
-
|
58 |
-
|
59 |
-
|
60 |
-
|
61 |
-
|
62 |
-
|
63 |
-
similarity = cosine_similarity(embedding1.detach().numpy(), embedding2.detach().numpy())
|
64 |
return similarity[0][0]
|
|
|
1 |
+
from transformers import CLIPProcessor, CLIPModel, ViTImageProcessor, ViTModel
|
2 |
+
from PIL import Image
|
3 |
+
from sklearn.metrics.pairwise import cosine_similarity
|
4 |
+
|
5 |
+
from warnings import filterwarnings
|
6 |
+
filterwarnings("ignore")
|
7 |
+
|
8 |
+
models = ["CLIP-ViT Base", "ViT Base", "DINO ViT-S16"]
|
9 |
+
models_info = {
|
10 |
+
"CLIP-ViT Base": {
|
11 |
+
"model_size": "386MB",
|
12 |
+
"model_url": "openai/clip-vit-base-patch32",
|
13 |
+
"efficiency": "High",
|
14 |
+
},
|
15 |
+
"ViT Base": {
|
16 |
+
"model_size": "304MB",
|
17 |
+
"model_url": "google/vit-base-patch16-224",
|
18 |
+
"efficiency": "High",
|
19 |
+
},
|
20 |
+
"DINO ViT-S16": {
|
21 |
+
"model_size": "1.34GB",
|
22 |
+
"model_url": "facebook/dino-vits16",
|
23 |
+
"efficiency": "Moderate",
|
24 |
+
},
|
25 |
+
}
|
26 |
+
|
27 |
+
class Image_Validator:
|
28 |
+
def __init__(self, model_name=None):
|
29 |
+
if model_name is None: model_name="ViT Base"
|
30 |
+
|
31 |
+
self.model_info = models_info[model_name]
|
32 |
+
model_url = self.model_info["model_url"]
|
33 |
+
|
34 |
+
if model_name == "CLIP-ViT Base":
|
35 |
+
self.model = CLIPModel.from_pretrained(model_url)
|
36 |
+
self.processor = CLIPProcessor.from_pretrained(model_url)
|
37 |
+
|
38 |
+
elif model_name == "ViT Base":
|
39 |
+
self.model = ViTModel.from_pretrained(model_url)
|
40 |
+
self.feature_extractor = ViTImageProcessor.from_pretrained(model_url)
|
41 |
+
|
42 |
+
elif model_name == "DINO ViT-S16":
|
43 |
+
self.model = ViTModel.from_pretrained(model_url)
|
44 |
+
self.feature_extractor = ViTImageProcessor.from_pretrained(model_url)
|
45 |
+
|
46 |
+
def get_image_embedding(self, image):
|
47 |
+
|
48 |
+
# Process image according to the model
|
49 |
+
if hasattr(self, 'processor'): # CLIP models
|
50 |
+
inputs = self.processor(images=image, return_tensors="pt")
|
51 |
+
outputs = self.model.get_image_features(**inputs)
|
52 |
+
|
53 |
+
elif hasattr(self, 'feature_extractor'): # ViT models
|
54 |
+
inputs = self.feature_extractor(images=image, return_tensors="pt")
|
55 |
+
outputs = self.model(**inputs).last_hidden_state
|
56 |
+
|
57 |
+
return outputs
|
58 |
+
|
59 |
+
def similarity_score(self, image1, image2):
|
60 |
+
embedding1 = self.get_image_embedding(image1).reshape(1, -1)
|
61 |
+
embedding2 = self.get_image_embedding(image2).reshape(1, -1)
|
62 |
+
similarity = cosine_similarity(embedding1.detach().numpy(), embedding2.detach().numpy())
|
|
|
63 |
return similarity[0][0]
|