forked from wmvanvliet/viswordrec-baseline
-
Notifications
You must be signed in to change notification settings - Fork 0
/
plot_pixel_contributions.py
132 lines (118 loc) · 4.32 KB
/
plot_pixel_contributions.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
"""
Use Captum to plot the pixel contributions to the output.
"""
import torch
import numpy as np
import pandas as pd
import mkl
mkl.set_num_threads(4)
from captum.attr import DeepLift
import matplotlib.pyplot as plt
from PIL import Image
from tqdm import tqdm
from torchvision import transforms
import network
import dataloader
# Path where you downloaded the OSF data to: https://osf.io/nu2ep/
data_path = './data'
# Whether to overwrite the existing figure
overwrite = False
classes = dataloader.TFRecord(f'{data_path}/training_datasets/epasana-10kwords').classes
classes.append(pd.Series(['noise'], index=[10000]))
# Load the TIFF images presented in the MEG experiment and apply the
# ImageNet preprocessing transformation to them.
stimuli = pd.read_csv(f'{data_path}/stimuli.csv')
preproc = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
])
images = []
for fname in tqdm(stimuli['tif_file'], desc='Reading images'):
with Image.open(f'{data_path}/stimulus_images/{fname}') as orig:
image = Image.new('RGB', (224, 224), '#696969')
image.paste(orig, (12, 62))
image = preproc(image).unsqueeze(0)
images.append(image)
images = torch.cat(images, 0)
# Load the model and feed through the images
checkpoint = torch.load(f'{data_path}/models/vgg11_first_imagenet_then_epasana-10kwords_noise.pth.tar', map_location='cpu')
model = network.VGG11.from_checkpoint(checkpoint, freeze=True)
model.eval()
model_output = model(images).detach().numpy()
layer_names = [
'conv1',
'conv1_relu',
'conv2',
'conv2_relu',
'conv3',
'conv3_relu',
'conv4',
'conv4_relu',
'conv5',
'conv5_relu',
'fc1',
'fc1_relu',
'fc2',
'fc2_relu',
'word',
'word_relu',
]
# Translate output of the model to text predictions
predictions = stimuli.copy()
predictions['predicted_text'] = classes[model_output.argmax(axis=1)].values
predictions['predicted_class'] = model_output.argmax(axis=1)
def plot_attributions(img, word=None, cl=None, rotation=0, ax=None, scale=.12):
if cl is None and word is not None:
cl = np.flatnonzero(classes == word)[0]
attributions = DeepLift(model).attribute(
img.unsqueeze(0), 0, target=int(cl),
return_convergence_delta=False)
attributions = attributions.detach().numpy().mean(axis=1)
if ax is None:
fig, ax = plt.subplots(1, 1)
vmax = scale
vmin = -vmax
ax.imshow(attributions[0][75:150], vmin=vmin, vmax=vmax, cmap='RdBu_r')
ax.set_axis_off()
print(np.abs(attributions).max())
if word is not None:
ax.set_title(f'{word}->{classes[cl]}', fontsize=10)
elif cl is not None:
ax.set_title(classes[cl], fontsize=10)
return ax
sel = predictions.query('type=="word"')
fig, axes = plt.subplots(15, 8, figsize=(14, 12))
for ax in axes.flat:
ax.set_axis_off()
for img, word, cl, ax in tqdm(zip(images[sel.index], sel.text, sel.predicted_class, axes.flat), total=len(sel)):
plot_attributions(img, word, ax=ax, cl=cl, scale=0.15)
plt.tight_layout()
sel = predictions.query('type=="pseudoword"')
fig, axes = plt.subplots(15, 8, figsize=(14, 12))
for ax in axes.flat:
ax.set_axis_off()
for img, word, cl, ax in tqdm(zip(images[sel.index], sel.text, sel.predicted_class, axes.flat), total=len(sel)):
plot_attributions(img, word, ax=ax, cl=cl, scale=0.06)
plt.tight_layout()
sel = predictions.query('type=="consonants"')
fig, axes = plt.subplots(15, 8, figsize=(14, 12))
for ax in axes.flat:
ax.set_axis_off()
for img, word, cl, ax in tqdm(zip(images[sel.index], sel.text, sel.predicted_class, axes.flat), total=len(sel)):
plot_attributions(img, word, ax=ax, cl=cl, scale=0.06)
plt.tight_layout()
sel = predictions.query('type=="symbols"')
fig, axes = plt.subplots(15, 8, figsize=(14, 12))
for ax in axes.flat:
ax.set_axis_off()
for img, cl, ax in tqdm(zip(images[sel.index], sel.predicted_class, axes.flat), total=len(sel)):
plot_attributions(img, ax=ax, cl=cl, scale=0.15)
plt.tight_layout()
sel = predictions.query('type=="noisy word"')
fig, axes = plt.subplots(15, 8, figsize=(14, 12))
for ax in axes.flat:
ax.set_axis_off()
for img, cl, ax in tqdm(zip(images[sel.index], sel.predicted_class, axes.flat), total=len(sel)):
plot_attributions(img, ax=ax, cl=cl, scale=0.06)
plt.tight_layout()