-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathblip_text_localization.py
136 lines (73 loc) · 2.94 KB
/
blip_text_localization.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
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
#!/usr/bin/env python
# coding: utf-8
# In[1]:
import torch
from PIL import Image
from lavis.models import load_model_and_preprocess
from lavis.processors import load_processor
# #### Load an example image and text
# In[2]:
file_path = r"C:\Users\kyanzhe\Downloads\blip2\image_path.txt" # Use raw string for file path
with open(file_path, 'r') as file:
data = file.read().rstrip().replace('"', '')
print(data)
img_path = data
raw_image = Image.open(img_path).convert('RGB') #doesnt work with .avif
# raw_image = Image.open("../docs/_static/merlion.png").convert("RGB")
# display(raw_image.resize((596, 437)))
# In[3]:
# setup device to use
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
# In[4]:
caption = "Merlion near marina bay."
# #### Load model and preprocessors
# In[5]:
# model, vis_processors, text_processors = load_model_and_preprocess("blip_image_text_matching", "base", device=device, is_eval=True)
model, vis_processors, text_processors = load_model_and_preprocess("blip_image_text_matching", "large", device=device, is_eval=True)
# #### Plot utilities for GradCam
# In[6]:
from matplotlib import pyplot as plt
from lavis.common.gradcam import getAttMap
from lavis.models.blip_models.blip_image_text_matching import compute_gradcam
import numpy as np
dst_w = 720
w, h = raw_image.size
scaling_factor = dst_w / w
resized_img = raw_image.resize((int(w * scaling_factor), int(h * scaling_factor)))
norm_img = np.float32(resized_img) / 255
# #### Preprocess image and text inputs
# In[7]:
img = vis_processors["eval"](raw_image).unsqueeze(0).to(device)
txt = text_processors["eval"](caption)
# #### Compute GradCam
# In[8]:
txt_tokens = model.tokenizer(txt, return_tensors="pt").to(device)
gradcam, _ = compute_gradcam(model, img, txt, txt_tokens, block_num=7)
# #### Average GradCam for the full image
# In[9]:
# avg_gradcam = getAttMap(norm_img, gradcam[0][1], blur=True)
gradcam_np = gradcam[0][1].numpy().astype(np.float32)
avg_gradcam = getAttMap(norm_img, gradcam_np, blur=True)
# In[10]:
# fig, ax = plt.subplots(num_image, 1, figsize=(15,5*num_image))
# fig, ax = plt.subplots(1, 1, figsize=(10, 10))
# ax.imshow(avg_gradcam)
# ax.show()
plt.imshow(avg_gradcam)
plt.show()
# #### GradCam for each token
# In[11]:
num_image = len(txt_tokens.input_ids[0]) - 2
fig, ax = plt.subplots(num_image, 1, figsize=(15, 5 * num_image))
gradcam_iter = iter(gradcam[0][2:-1])
token_id_iter = iter(txt_tokens.input_ids[0][1:-1])
np.seterr(divide='ignore', invalid='ignore') #Divide by zero TODO
for i, (gradcam, token_id) in enumerate(zip(gradcam_iter, token_id_iter)):
word = model.tokenizer.decode([token_id])
# gradcam_image = getAttMap(norm_img, gradcam, blur=True)
gradcam_np = gradcam[0][1].numpy().astype(np.float32)
gradcam_image = getAttMap(norm_img, gradcam_np, blur=True)
ax[i].imshow(gradcam_image)
ax[i].set_yticks([])
ax[i].set_xticks([])
ax[i].set_xlabel(word)