-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathvis.py
180 lines (151 loc) · 6.86 KB
/
vis.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
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
import clip
import torch
import torch.nn.functional as F
from pascal_5i_loader import Pascal5iLoader
from config import config
from torch.utils.data import DataLoader
from tqdm import tqdm
from torch.utils.tensorboard import SummaryWriter
import sys
import json
import matplotlib.pyplot as plt
import numpy as np
# previous_runs = os.listdir('images')
# if len(previous_runs) == 0:
# run_number = 1
# else:
# run_number = max([int(s.split('run_')[1]) for s in previous_runs]) + 1
config['runid'] = int(sys.argv[1])
logdir = f"run_{config['runid']}"
run = json.load(open('fewshotruns.json'))[f"{config['runid']}"]
config['fold'] = run['fold']
config['img_size'] = run['img_size']
config['model_name'] = run['model_name']
writer = SummaryWriter('fsimages/'+logdir)
def norm_im(im):
x_min, x_max = im.min(), im.max()
ims = (im - x_min) / (x_max-x_min)
return ims
def heatmap(data, row_labels, col_labels, ax=None, cbar_kw={}, cbarlabel="", **kwargs):
fig = plt.figure()
# Plot the heatmap
plt.imshow(data, **kwargs)
# Create colorbar
plt.colorbar(**cbar_kw)
plt.ylabel(cbarlabel, rotation=-90, va="bottom")
# Show all ticks and label them with the respective list entries.
plt.xticks(np.arange(data.shape[1]), labels=col_labels)
plt.yticks(np.arange(data.shape[0]), labels=row_labels)
# Let the horizontal axes labeling appear on top.
plt.tick_params(top=True, bottom=False, labeltop=True, labelbottom=False)
# Turn spines off and create white grid.
# plt.spines[:].set_visible(False)
plt.box(False)
plt.setp(plt.xticks()[1], rotation=-90, ha="right", rotation_mode="anchor")
# plt.xticks(np.arange(data.shape[1]+1)-.5, minor=True)
# plt.yticks(np.arange(data.shape[0]+1)-.5, minor=True)
plt.grid(which="minor", color="w", linestyle='-', linewidth=3)
plt.tick_params(which="minor", bottom=False, left=False)
return fig
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
print('Running on', device, 'logging in', logdir)
if config['model_name'] == 'CLIP':
from models import model_orig
model, preproc, preproc_lbl = model_orig.load_custom_clip('RN50', device=device, img_size=config['img_size'])
elif config['model_name'] == 'PSPNet':
from models import model_pspnet
model, preproc = model_pspnet.load_segclip_psp(zoom=config['zoom'], img_size=config['img_size'], device=device)
preproc_lbl = None
model.to(device) # redundant
pretrained_dict = torch.load(f"fewshotruns/run_{config['runid']}/model.pt", map_location=device)
pretrained_dict = {k: v for k, v in pretrained_dict.items() if k in model.state_dict()}
model.load_state_dict(pretrained_dict)
# dataset = pascalVOCLoader(config['pascal_root'], preproc, preproc_lbl, split='train', img_size=224, is_transform=True)
# trainloader = DataLoader(dataset, batch_size=config['batch_size'], pin_memory=True, num_workers=config['num_workers'])
dataset = Pascal5iLoader(config['pascal_root'], fold=config['fold'], preproc=preproc, preproc_lbl=preproc_lbl)
trainloader = DataLoader(dataset, batch_size=config['batch_size'], pin_memory=True, num_workers=config['num_workers'])
valset = Pascal5iLoader(config['pascal_root'], fold=config['fold'], preproc=preproc, preproc_lbl=preproc_lbl, train=False)
valloader = DataLoader(valset, batch_size=config['batch_size'], pin_memory=True, num_workers=config['num_workers'])
pascal_classes = [
'aeroplane',
'bicycle',
'bird',
'boat',
'bottle',
'bus',
'car',
'cat',
'chair',
'cow',
'dog',
'horse',
'motorbike',
'person',
'sheep',
'sofa',
'diningtable',
'pottedplant',
'train',
'tvmonitor',
]
template = 'a photo of a '
pascal_labels = [template+x for x in pascal_classes]
pascal_labels.insert(0, '')
pascal_labels_train = [pascal_labels[x] for x in dataset.label_set]
# pascal_labels_val = [pascal_labels[x] for x in valset.label_set]
pascal_labels_val = pascal_labels
pascal_labels_train.insert(0, '')
# pascal_labels_val.insert(0, '')
text_tokens_train = clip.tokenize(pascal_labels_train).to(device)
text_tokens_val = clip.tokenize(pascal_labels_val).to(device)
print(text_tokens_train.shape)
print(text_tokens_val.shape)
# with profile(activities=[ProfilerActivity.CPU, ProfilerActivity.CUDA], record_shapes=True) as prof:
# with record_function("model_inference"):
model.eval()
# approx ones only do +1 instead of +(num_of_pixels) for a pair of (gt, pred) labels
confusion_matrix = torch.zeros(21,21, device=device)
confusion_matrix_approx = torch.zeros(21,21, device=device)
val_confusion_matrix = torch.zeros(21,21, device=device)
val_confusion_matrix_approx = torch.zeros(21,21, device=device)
for i,(img,lbl) in tqdm(enumerate(trainloader), total=len(trainloader)):
img, lbl = img.to(device), lbl.to(device)
pred = model(img, text_tokens_train)
pred = F.softmax(pred, dim=1).argmax(dim=1)
if i < 20:
pred = torch.stack([valset.decode_segmap(x).permute(2,0,1) for x in pred]).to(device)
lbl = torch.stack([valset.decode_segmap(x).permute(2,0,1) for x in lbl]).to(device)
writer.add_images('TRAIN img vs pred vs GT', torch.cat([norm_im(img), norm_im(pred), norm_im(lbl)], dim=2), global_step=i)
lbl, pred = lbl.long(), pred.long().to(device)
for label in torch.unique(lbl):
if label == 255:
continue
pred_classes, counts = torch.unique(pred[lbl==label], return_counts=True)
for j in range(len(pred_classes)):
confusion_matrix[label][pred_classes[j]] += counts[j]
confusion_matrix_approx[label][pred_classes[j]] += 1
torch.save(confusion_matrix, 'fsimages/'+logdir+'/conf.pt')
writer.add_figure('conf', heatmap(confusion_matrix.cpu()[1:,1:], pascal_classes, pascal_classes))
torch.save(confusion_matrix_approx, 'fsimages/'+logdir+'/conf_approx.pt')
writer.add_figure('conf_approx', heatmap(confusion_matrix_approx.cpu()[1:,1:], pascal_classes, pascal_classes))
for i,(img,lbl) in tqdm(enumerate(valloader), total=len(valloader)):
img, lbl = img.to(device), lbl.to(device)
pred = model(img, text_tokens_val)
pred = F.softmax(pred, dim=1).argmax(dim=1)
if i == 20:
pred = torch.stack([valset.decode_segmap(x).permute(2,0,1) for x in pred]).to(device)
lbl = torch.stack([valset.decode_segmap(x).permute(2,0,1) for x in lbl]).to(device)
writer.add_images('VAL img vs pred vs GT', torch.cat([norm_im(img), norm_im(pred), norm_im(lbl)], dim=2), global_step=i)
lbl, pred = lbl.long(), pred.long().to(device)
for label in torch.unique(lbl):
if label == 255:
continue
pred_classes, counts = torch.unique(pred[lbl==label], return_counts=True)
for j in range(len(pred_classes)):
val_confusion_matrix[label][pred_classes[j]] += counts[j]
val_confusion_matrix_approx[label][pred_classes[j]] += 1
torch.save(val_confusion_matrix, 'fsimages/'+logdir+'/val_conf.pt')
writer.add_figure('val_conf', heatmap(val_confusion_matrix.cpu()[1:,1:], pascal_classes, pascal_classes))
torch.save(val_confusion_matrix_approx, 'fsimages/'+logdir+'/val_conf_approx.pt')
writer.add_figure('val_conf_approx', heatmap(val_confusion_matrix_approx.cpu()[1:,1:], pascal_classes, pascal_classes))
writer.close()