-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathvis.py
339 lines (242 loc) · 11.6 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
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
import torch
import torch.nn as nn
import os
import argparse
from utils.logger import setup_logger
from utils.input import frames_preprocess
from visualizer import get_local
get_local.activate() # 激活装饰器
from models.model import CAT
from train import setup_seed
from data.dataset import load_dataset
from configs.defaults import get_cfg_defaults
import pdb
from tqdm import tqdm
import cv2
import numpy as np
import torch.nn.functional as F
def load_model(model_path):
model = CAT(num_class=cfg.DATASET.NUM_CLASS,
num_clip=cfg.DATASET.NUM_CLIP,
dim_embedding=cfg.MODEL.DIM_EMBEDDING,
backbone_model=cfg.MODEL.BACKBONE,
backbone_dim=cfg.MODEL.BACKBONE_DIM,
base_model=cfg.MODEL.BASE_MODEL,
pretrain=cfg.MODEL.PRETRAIN,
dropout=cfg.TRAIN.DROPOUT,
use_ViT=cfg.MODEL.TRANSFORMER,
use_SeqAlign=cfg.MODEL.ALIGNMENT,
use_CosFace=cfg.MODEL.COSFACE).to(device)
checkpoint = torch.load(model_path)
model.load_state_dict(checkpoint['model_state_dict'], strict=True)
print('Load state dict form %s' % model_path)
return model
def vis_embedding(model_path):
id_bank = ['1.1', '1.2', '1.3', '1.4', '1.5']
# id_bank = ['5.1', '5.2', '5.3', '5.4', '5.5']
# id_bank = ['18.1', '18.3', '18.4', '18.5', '18.6', '18.7', '18.13', '18.15', '18.16', '18.18']
model = load_model(model_path)
if torch.cuda.device_count() > 1 and torch.cuda.is_available():
# logger.info("Let's use %d GPUs" % torch.cuda.device_count())
model = torch.nn.DataParallel(model)
labels = []
embeds = []
with torch.no_grad():
for iter, sample in enumerate(tqdm(test_loader)):
# print(sample['index'])
# indices += sample['index']
# continue
frames_list1 = sample['frames_list1']
# frames_list2 = sample['frames_list2']
# assert len(frames_list1) == len(frames_list2)
# pdb.set_trace()
labels1 = [id_bank.index(tmp) for tmp in sample['label1']]
# labels2 = [id_bank.index(tmp) for tmp in sample['label2']]
# label = labels1+labels2
labels += labels1
embed1s = []
# embed2s = []
# num_true_pred = 0
# pdb.set_trace()
for i in range(len(frames_list1)):
frames1 = frames_preprocess(frames_list1[i], cfg.MODEL.BACKBONE_DIM, cfg.MODEL.BACKBONE).to(device, non_blocking=True)
# frames2 = frames_preprocess(frames_list2[i], cfg.MODEL.BACKBONE_DIM, cfg.MODEL.BACKBONE).to(device, non_blocking=True)
# pdb.set_trace()
pred1, seq_features1, embed1 = model(frames1)
# pred2, seq_features2, embed2 = model(frames2)
# true_labels1 = torch.tensor([LABELS['COIN']['train'].index(label_str) for label_str in labels1]).to(device)
# true_labels2 = torch.tensor([LABELS['COIN']['train'].index(label_str) for label_str in labels2]).to(device)
#
# pred_labels1 = torch.argmax(tmp1, dim=-1)
# pred_labels2 = torch.argmax(tmp2, dim=-1)
# num_true_pred = torch.sum(pred_labels1 == true_labels1) + torch.sum(pred_labels2 == true_labels2)
# print('Accuracy: %4f' % (num_true_pred / (len(true_labels1) + len(true_labels2))))
# # pdb.set_trace()
#
# pred1 = model(frames1, embed=True)
# pred2 = model(frames2, embed=True)
#
# distance = torch.sum((pred1 - pred2) ** 2, dim=1)
embed1s.append(embed1)
# embed2s.append(embed2)
# pdb.set_trace()
# pdb.set_trace()
embed1s = np.sum(embed1s) / len(embed1s)
# embed2s = np.sum(embed2s) / len(embed2s)
embeds.append(embed1s)
# embeds.append(embed2s)
# pdb.set_trace()
embeds = torch.cat(embeds)
embeds1, embeds2, embeds3 = 0, 0, 0
for i in range(len(labels)):
if labels[i] == 2:
embeds1 += embeds[i]
if labels[i] == 3:
embeds2 += embeds[i]
if labels[i] == 4:
embeds3 += embeds[i]
labels = torch.tensor(labels)
# embeds = torch.cat([embeds, (embeds1/41).unsqueeze(0), (embeds2/41).unsqueeze(0), (embeds3/41).unsqueeze(0)], dim=0)
# labels = torch.tensor(labels + [7,8,9])
# PCA
from sklearn.decomposition import PCA
import matplotlib.pyplot as plt
embeds = F.normalize(embeds, p=2, dim=1)
pca_sk = PCA(n_components=2)
coords = pca_sk.fit_transform(np.array(embeds.cpu()))
coords0 = coords[labels == 0]
coords1 = coords[labels == 1]
coords2 = coords[labels == 2]
colors0 = ['c' for i in coords0]
colors1 = ['y' for i in coords1]
colors2 = ['g' for i in coords2]
center0 = [coords0[:, 0].mean(), coords0[:, 1].mean()]
center1 = [coords1[:, 0].mean(), coords1[:, 1].mean()]
center2 = [coords2[:, 0].mean(), coords2[:, 1].mean()]
coords = np.concatenate((coords0, coords1, coords2), axis=0)
centers = np.concatenate(([center0], [center1], [center2]), axis=0)
print('Variance: 1.1 - %.4f; 1.2 - %.4f; 1.3 - %.4f; centers - %.4f' %
((np.var(coords0[:,0])+np.var(coords0[:,1])) / 2,
(np.var(coords1[:,0])+np.var(coords1[:,1])) / 2,
(np.var(coords2[:,0])+np.var(coords2[:,1])) / 2,
(np.var(centers[:,0])+np.var(centers[:,1])) / 2))
plt.scatter(coords[:, 0], coords[:, 1], c=colors0+colors1+colors2, marker='o')
# plt.scatter(centers[:, 0], centers[:, 1], c=['b', 'y', 'r'], marker='x')
plt.savefig('vis1.png')
pdb.set_trace()
# Tensorboard visualization
img = cv2.resize(cv2.imread('square.png'), (30, 30), interpolation=cv2.INTER_AREA)
img = torch.tensor(img).unsqueeze(0).repeat(embeds.size(0), 1, 1, 1).permute(0, 3, 1, 2)
pdb.set_trace()
from tensorboardX import SummaryWriter
# from torch.utils.tensorboard import SummaryWriter
writer = SummaryWriter('runs/exp_123_new')
# writer.add_embedding(embeds, labels, global_step=0)
writer.add_embedding(embeds, labels, img, global_step=0)
def process_one_epoch(model_path, vis_func, **kwargs):
model = load_model(model_path)
if torch.cuda.device_count() > 1 and torch.cuda.is_available():
# logger.info("Let's use %d GPUs" % torch.cuda.device_count())
model = torch.nn.DataParallel(model)
# model = model.to(device)
auc_value = 0
indices = []
# auc metric
model.eval()
with torch.no_grad():
for iter, sample in enumerate(tqdm(test_loader)):
# print(sample['index'])
# indices += sample['index']
# continue
frames_list1 = sample['frames_list1']
frames_list2 = sample['frames_list2']
assert len(frames_list1) == len(frames_list2)
labels1 = sample['label1']
labels2 = sample['label2']
label = torch.tensor(np.array(labels1) == np.array(labels2)).to(device)
pred1 = 0
pred2 = 0
# pdb.set_trace()
for i in range(len(frames_list1)):
frames1 = frames_preprocess(frames_list1[i], cfg.MODEL.BACKBONE_DIM, cfg.MODEL.BACKBONE).to(device, non_blocking=True)
frames2 = frames_preprocess(frames_list2[i], cfg.MODEL.BACKBONE_DIM, cfg.MODEL.BACKBONE).to(device, non_blocking=True)
# pdb.set_trace()
pred1 += model(frames1, embed=True)
pred2 += model(frames2, embed=True)
attn1 = get_local.cache
vis_func(frames_preprocess(sample["raw_frames_list1"][i], cfg.MODEL.BACKBONE_DIM, cfg.MODEL.BACKBONE), attn1['Attention.forward'])
# pdb.set_trace()
pred1 /= len(frames_list1)
pred2 /= len(frames_list1)
# L1 distance
# pred = torch.sum(torch.abs(pred1 - pred2), dim=1)
# L2 distance
pred = torch.sum((pred1 - pred2) ** 2, dim=1)
if iter == 0:
preds = pred
labels = label
else:
preds = torch.cat([preds, pred])
labels = torch.cat([labels, label])
def vis_attn(frames, attns):
pdb.set_trace()
bs, _, h, w = frames.size()
frames = frames.reshape(cfg.TRAIN.BATCH_SIZE, cfg.DATASET.NUM_CLIP, 3, h, w)
attns = [torch.diagonal(torch.from_numpy(attn),dim1=2,dim2=3).mean(dim=1, keepdim=True)[:,:,1:] for attn in attns] # [[bs, 1, H*W],...]
def attn_seq2map(attn_seq, intermediate_shape, output_shape):
# pdb.set_trace()
bs, c, _ = attn_seq.size()
h, w = intermediate_shape
length = h * w
upsampler = nn.Upsample(size=output_shape, mode='bilinear')
attn_maps = []
for i in range(int(attn_seq.size(-1) / length)):
seq = attn_seq[:,:,i*length:(i+1)*length]
scale = 1 / seq.max()
seq = seq * scale
seq = seq.reshape(bs, c, h, w)
attn_map =upsampler(seq)
attn_maps.append(attn_map.unsqueeze(1))
return torch.cat(attn_maps, dim=1)
# return torch.cat([upsampler((]*scale).reshape(bs, c, h, w)).unsqueeze(1) for i in range(int(attn_seq.size(-1)/length))], dim=1)
def save_heatmap_img(img, mask, path):
# pdb.set_trace()
heatmap = cv2.applyColorMap(np.uint8(255 * mask), cv2.COLORMAP_JET)
heatmap = np.float32(heatmap) / 255
img = cv2.cvtColor(np.asarray(img), cv2.COLOR_RGB2BGR)
cam = 0.4 * heatmap + 0.6 * np.float32(img)
cam = cam / np.max(cam)
cv2.imwrite(path, np.uint8(255 * cam))
# cam = cam[:, :, ::-1] # BGR > RGB
# plt.figure(figsize=(10, 10))
# plt.imshow(np.uint8(255 * cam))
for i in range(cfg.TRAIN.BATCH_SIZE):
for k in range(len(attns)):
attn_maps = attn_seq2map(attns[k], (6, 10), (180, 320)) # [bs, 16, 1, 180, 320]
for j in range(cfg.DATASET.NUM_CLIP):
# pdb.set_trace()
split_list = args.model_path.split('/')
save_path = os.path.join('figs', split_list[1], split_list[3][:-4])
if not os.path.exists(save_path):
os.makedirs(save_path)
save_path = os.path.join(save_path, 'attn_bs%d_layers%d_idx%d.jpg' % (i+1, k+1, j+1))
save_heatmap_img(frames[i, j].permute(1, 2, 0), attn_maps[i, j].permute(1, 2, 0), save_path)
pdb.set_trace()
def parse_args():
parser = argparse.ArgumentParser()
parser.add_argument('--config', default='configs/eval_resnet_config.yml', help='config file path [default: configs/test_config.yml]')
parser.add_argument('--model_path', default=None, help='path to load one model [default: None]')
# parser.add_argument('--log_name', default='eval_log', help='log name')
args = parser.parse_args()
return args
if __name__ == "__main__":
args = parse_args()
cfg = get_cfg_defaults()
if args.config:
cfg.merge_from_file(args.config)
setup_seed(cfg.TRAIN.SEED)
use_cuda = cfg.TRAIN.USE_CUDA and torch.cuda.is_available()
device = torch.device("cuda" if use_cuda else "cpu")
test_loader = load_dataset(cfg)
vis_embedding(args.model_path)
# process_one_epoch(args.model_path, vis_attn)