-
Notifications
You must be signed in to change notification settings - Fork 0
/
feature_extraction_cifar10.py
74 lines (62 loc) · 2.54 KB
/
feature_extraction_cifar10.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
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
import torch
import numpy as np
import os
from models.backbone_attention import Backbone
import torchvision
from torchvision.io import read_image
from torchvision.io.image import ImageReadMode
from imutils.video import fps
from sklearn.metrics.pairwise import cosine_distances
from matplotlib import pyplot as plt
import cv2
from torch.utils.data import Dataset, DataLoader
from utils.utils import normalize
if __name__ == '__main__':
os.environ["CUDA_VISIBLE_DEVICES"] = "1"
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model_path = "./results/model_cifar10_SupervisedContrastiveLoss.pth"
backbone = Backbone(out_dimension=256, model_name="resnet18", pretrained=False)
model, _, _ = backbone.build_model()
model.load_state_dict(torch.load(model_path))
model.to(device)
model.eval()
# print(model)
val_transforms_list = [
torchvision.transforms.Resize(size=(224, 224)),
torchvision.transforms.ToTensor(),
torchvision.transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
]
val_transforms = torchvision.transforms.Compose(val_transforms_list)
val_dataset = torchvision.datasets.CIFAR10(root="./datasets", train=False, transform=val_transforms, download=True)
classes = val_dataset.classes
val_data_loader = DataLoader(val_dataset, batch_size=1, drop_last=False, shuffle=False, num_workers=4)
feature_data = {}
cls_idx_list = []
count = 0
for data, cls_idx in val_data_loader:
cls_idx = cls_idx.item()
if cls_idx not in feature_data:
feature_data[cls_idx] = []
if cls_idx in feature_data and len(feature_data[cls_idx]) == 10:
continue
feature = model(data.cuda())
feature = normalize(feature, dim=1)
feature = feature.cpu().detach().numpy()
print("feature: {}".format(feature.shape))
feature_data[cls_idx].append(feature)
count += 1
print(count, cls_idx)
# print(list(feature_data.values()))
feature_data = np.concatenate(list(feature_data.values()), axis=0).squeeze()
print(feature_data.shape)
print(np.linalg.norm(feature_data, ord=2, axis=1))
distance = cosine_distances(feature_data, feature_data)
similarity = 1.0 - distance
print(similarity)
plt.figure(figsize=(50, 50))
plt.imshow(similarity)
for i in range(similarity.shape[0]):
for j in range(similarity.shape[1]):
plt.text(j, i, "{:.1f}".format(similarity[i, j]), ha='center', va='center', color='w')
plt.savefig("./results/plot_cifar10_with_value.jpg")
plt.show()