-
Notifications
You must be signed in to change notification settings - Fork 0
/
semantic_score.py
52 lines (42 loc) · 1.54 KB
/
semantic_score.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
import os
import torch
import clip
from tqdm import tqdm
from PIL import Image
from typing import List
import pandas as pd
def embed_images(image_paths: List[str]) -> torch.Tensor:
"""
Embeds a list of image paths using CLIP.
Args:
image_paths (List[str]): A list of image paths.
Returns:
torch.Tensor: A tensor containing the image embeddings.
"""
torch.manual_seed(0)
device = "cuda" if torch.cuda.is_available() else "cpu"
model, preprocess = clip.load("ViT-B/32", device=device)
images = [preprocess(Image.open(path)).unsqueeze(0) for path in image_paths]
images = torch.cat(images, dim=0).to(device)
with torch.no_grad():
embeddings = model.encode_image(images)
return embeddings
def calculate_similarity(embeddings: torch.Tensor) -> float:
"""
Calculates the mean pairwise cosine similarity between image embeddings.
Args:
embeddings (torch.Tensor): A tensor containing the image embeddings.
Returns:
float: The mean pairwise cosine similarity scaled between 0 and 100.
"""
cosine_similarities = []
for i in tqdm(range(len(embeddings)), desc="Calculating pairwise similarities"):
for j in range(i + 1, len(embeddings)):
cos_sim = torch.cosine_similarity(embeddings[i], embeddings[j], dim=0)
scaled_similarity = max(100 * cos_sim.item(), 0)
cosine_similarities.append(scaled_similarity)
return torch.tensor(cosine_similarities).mean().item()
def main():
pass
if __name__ == "__main__":
main()