MastiDiff/test_label.py at main · MasterToycode/MastiDiff · GitHub

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
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
import torch
import torch.nn as nn
import numpy as np
import matplotlib.pyplot as plt

from sklearn.manifold import TSNE
from sklearn.decomposition import PCA
from sklearn.metrics.pairwise import cosine_similarity

from diffusers import UNet2DConditionModel
from train_ddpm_variance import LabelEmbedding

# ===============================
# 配置
# ===============================

checkpoint_dir = r"ddpm_variance_10\checkpoint_epoch_99"

num_classes = 4
uncond_label = 4

device = "cuda" if torch.cuda.is_available() else "cpu"

# ===============================
# 加载模型
# ===============================

model = UNet2DConditionModel.from_pretrained(checkpoint_dir).to(device).eval()

label_proj = LabelEmbedding(num_classes + 1).to(device).eval()

try:
    label_proj.load_state_dict(
        torch.load(f"{checkpoint_dir}/label_proj.pt", map_location=device)
    )
    print("✅ label_proj loaded")
except:
    print("⚠️ label_proj 未找到权重，使用随机初始化")


# ===============================
# 提取 embedding
# ===============================

with torch.no_grad():

    labels = torch.arange(num_classes + 1).to(device)

    tokens = label_proj(labels)   # (C, 8, 512)

    emb_mean = tokens.mean(dim=1)  # (C,512)

    emb_flat = tokens.view(num_classes + 1, -1)  # (C,4096)

emb_mean = emb_mean.cpu().numpy()
emb_flat = emb_flat.cpu().numpy()

# ===============================
# 1 t-SNE
# ===============================

print("Running t-SNE...")

tsne = TSNE(n_components=2, perplexity=3, random_state=42)
emb_2d = tsne.fit_transform(emb_mean)

plt.figure(figsize=(6,6))
plt.scatter(emb_2d[:,0], emb_2d[:,1],
            c=np.arange(num_classes+1),
            cmap="tab10",
            s=150)

for i in range(num_classes+1):
    plt.text(emb_2d[i,0], emb_2d[i,1], f"class {i}")

plt.title("Label Embedding t-SNE")
plt.show()


# ===============================
# 2 距离矩阵
# ===============================

print("Embedding distance matrix:")

emb_torch = torch.tensor(emb_flat)

dist = torch.cdist(emb_torch, emb_torch)

print(dist)

plt.imshow(dist.numpy())
plt.colorbar()
plt.title("Embedding Distance Matrix")
plt.show()


# ===============================
# 3 Cosine similarity
# ===============================

cos = cosine_similarity(emb_flat)

print("Cosine similarity:")

print(cos)

plt.imshow(cos)
plt.colorbar()
plt.title("Cosine Similarity Matrix")
plt.show()


# ===============================
# 4 Token 结构
# ===============================

tokens_np = tokens.cpu().numpy()

plt.figure(figsize=(8,4))

for c in range(num_classes+1):

    token_norm = np.linalg.norm(tokens_np[c], axis=1)

    plt.plot(token_norm, label=f"class {c}")

plt.legend()
plt.title("Token Norm Distribution")
plt.xlabel("Token Index")
plt.ylabel("Norm")
plt.show()


# ===============================
# 5 PCA
# ===============================

pca = PCA(n_components=2)

pca_2d = pca.fit_transform(emb_flat)

plt.figure(figsize=(6,6))

plt.scatter(pca_2d[:,0], pca_2d[:,1],
            c=np.arange(num_classes+1),
            cmap="tab10",
            s=150)

for i in range(num_classes+1):
    plt.text(pca_2d[i,0], pca_2d[i,1], f"class {i}")

plt.title("Label Embedding PCA")
plt.show()


print("Explained variance ratio:", pca.explained_variance_ratio_)