-
Notifications
You must be signed in to change notification settings - Fork 0
/
test.py
339 lines (290 loc) · 13 KB
/
test.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
339
import re
import glob
from operator import rshift
import torch
import torch.nn as nn
from torch.utils.data import DataLoader
from torch.autograd import Variable
import torch.nn.functional as F
import argparse
import os
import json
import pickle
import utils
import numpy as np
import pandas as pd
from tqdm import tqdm
import matplotlib.pyplot as plt
from mpl_toolkits.mplot3d import Axes3D
from matplotlib import patches
import seaborn as sns
from sklearn.manifold import TSNE
from sklearn.decomposition import PCA
from scipy import interpolate
from scipy.stats import chi2, norm
import copy
import shutil
from transformers import BertModel, BertConfig, BertTokenizer
from model import ACT2G
# from model_wokf import ACT2G
# from model_20220907 import ACT2G
from plotPose import Plot
from option import getOptions
opt = getOptions()
def reorder(sequence): # ([128, 8, 64, 64, 3])
return sequence.permute(0,1,3,2)
# os.environ['CUDA_VISIBLE_DEVICES'] = opt.gpu
os.environ['KMP_DUPLICATE_LIB_OK']='True'
def plotUpperBody2D(pose3d, save_path, fps=25, isRelative=False, frames=None):
if isRelative:
new_pose3d = []
for pose in pose3d:
new_pose = np.zeros([8, 3])
new_pose[0] = pose[0]
new_pose[2] = pose[1]
new_pose[3] = new_pose[2] + pose[2]
new_pose[4] = new_pose[3] + pose[3]
new_pose[5] = pose[4]
new_pose[6] = new_pose[5] + pose[5]
new_pose[7] = new_pose[6] + pose[6]
new_pose3d.append(new_pose)
pose3d = np.array(new_pose3d)
if pose3d.shape[1] == 25:
upper_idx = [2, 20, 4, 5, 6, 8, 9, 10]
pose3d = pose3d[:, upper_idx]
pose2d = np.delete(pose3d, 2, 2).reshape([-1, pose3d.shape[1]*2])
else:
pose2d = np.delete(pose3d, 2, 2).reshape([-1, pose3d.shape[1]*2])
if frames is not None:
pose2d = np.append(pose2d.T, [frames], axis=0).T
p = Plot((-0.5, 0.5), (-0.6, 0.4))
# p = Plot((-0.5, 0.5), (-0.4, 0.6))
anim = p.animate(pose2d, 1000/fps)
p.save(anim, save_path, fps=fps)
def plotResult(X, labels, text=None):
unique_labels = set(labels)
cmap = plt.rcParams['axes.prop_cycle'].by_key()['color']
colors = [cmap[i%len(cmap)] for i in range(len(labels))]
X_reduced = TSNE(n_components=2, random_state=0).fit_transform(X)
for k, col in zip(unique_labels, colors):
class_member_mask = labels == k
xy = X_reduced[class_member_mask]
if text is not None:
texts = np.array(text)[class_member_mask]
for i in range(0, len(xy), 1):
plt.text(xy[i][0], xy[i][1], texts[i], color=col)
plt.xlim(-40, 70)
plt.ylim(-45, 45)
else:
plt.scatter(xy[:, 0], xy[:, 1], marker="${}$".format(k), label=k, color=col)
plt.title("KMeans Clustering")
plt.show()
def visualizePrediction(tokens, pred, labels=False, save_path=None):
fig, ax = plt.subplots()
idx = tokens.index('[PAD]')
# idx = len(tokens)
if labels is not None:
data = np.array([pred[:idx], labels[:idx]]).T
else:
data = np.array([pred[:idx]]).T
sns.heatmap(data=data, cmap='OrRd', annot=True, vmin=0, vmax=1)
ax.set_xticklabels(['Predicted', 'GT'], rotation=0)
ax.set_yticks(np.arange(idx)+0.5)
ax.set_yticklabels(tokens[:idx], rotation=0)
plt.tight_layout()
if save_path:
plt.savefig(save_path)
plt.close()
else:
plt.show()
def interpolateKeypose(keyposes, frame=60): # [keyframe, joint, xyz]
keyposes = keyposes.transpose([1, 2, 0])
new_poses = []
for pose in keyposes:
tck, u = interpolate.splprep(pose, k=3)
new = interpolate.splev(np.linspace(0,1,frame), tck, der=0)
new_poses.append(new)
new_poses = np.array(new_poses).transpose([2, 0, 1])
return new_poses
def main(opt):
if opt.model_path != '':
model_path = opt.model_path
saved_model = torch.load(model_path)
opt = saved_model['option']
opt.batch_size = 256
opt.model_path = model_path
opt.mode = "test"
model = ACT2G(opt).cuda()
model.load_state_dict(saved_model['model'])
else:
raise ValueError('missing checkpoint')
# BERT setting
options_name = "bert-base-uncased"
tokenizer = BertTokenizer.from_pretrained(options_name)
bert_model = BertModel.from_pretrained(options_name)
bert_model.eval()
device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')
print("Device : ", device)
model = model.to(device)
model.eval()
# --------- load a dataset ------------------------------------
# train_data, test_data = utils.load_dataset_textreconst(opt)
train_data, test_data, positive = utils.load_dataset(opt)
train_loader = DataLoader(train_data,
num_workers=0,
batch_size=opt.batch_size,
shuffle=False,
drop_last=True,
pin_memory=True)
test_loader = DataLoader(test_data,
num_workers=0,
batch_size=opt.batch_size,
shuffle=True,
drop_last=True,
pin_memory=True)
# # save gesture images
# out = pickle.load(open(opt.dataset, "rb"))
# gestures = out['gestures']
# texts = out['remarks']
# save_dir = './images/Gesture/gestures/stick_1002_key/'
# os.makedirs(save_dir, exist_ok=True)
# for i in range(len(gestures)):
# text = re.sub(r'[\\/:*?"<>|]+','',texts[i][:30])
# save_path = save_dir + str(i).zfill(4) + '_{}.gif'.format(text)
# if os.path.exists(save_path):
# continue
# frames = np.arange(len(gestures[i]))
# plotUpperBody2D(gestures[i], save_path, fps=4, isRelative=False, frames=frames)
# exit()
# --------- training loop ------------------------------------
# save_dir = "./images/Gesture/reconst_train_{}keyframe_1000ges_r-1000_f=256dim_linear/".format(opt.frames)
save_dir = "./images/reconst/{}/".format(os.path.basename(os.path.dirname(opt.model_path)))
os.makedirs(save_dir, exist_ok=True)
os.makedirs(save_dir+'gesture/', exist_ok=True)
os.makedirs(save_dir+'gesture_allframe/', exist_ok=True)
os.makedirs(save_dir+'gesture_interp/', exist_ok=True)
os.makedirs(save_dir+'attention/', exist_ok=True)
os.makedirs(save_dir+'attention_norm/', exist_ok=True)
isRelative = True
# isRelative = False
kf_gestures, recon_gestures, g_features, t_features, attns, kfs, kf_preds = [], [], [], [], [], [], []
keyframes = []
for i, data in enumerate(train_loader):
x_gesture, x_text, y_text, kf_list = data['gesture'].cuda(), data['text'].cuda(), data['text_label'].cuda(), data['kf_list'].cuda()
x_gesture = x_gesture.type(torch.float32).to(device)
if opt.batch_size != 1:
pos = positive[data['index']][:,data['index']]
recon_g, gesture_feature, text_feature, attn, kf_pred = model(x_gesture, x_text)
# recon_g, gesture_feature, text_feature, attn = model(x_gesture, x_text)
# kf_pred = torch.argmax(kf_pred, dim=1).detach().cpu().numpy()
g_features.append(gesture_feature.detach().cpu().numpy()[0])
t_features.append(text_feature.detach().cpu().numpy()[0])
kf_gestures.append(data['gesture'].detach().cpu().numpy()[0])
recon_gestures.append(recon_g.detach().cpu().numpy()[0])
attns.append(attn.detach().cpu().numpy()[0])
kfs.append(kf_list.detach().cpu().numpy()[0])
# kf_preds.append(kf_pred[0])
kf_list = kf_list.detach().cpu().numpy()
# df = pd.DataFrame({"GT": kf_list, "Pred": kf_pred})
# df.to_excel(save_dir+"keyframe_pred.xlsx", index=False)
############ Reconst Gestures ############
sep = 1 if opt.frames > 30 else 0
fps = 25 if opt.frames > 30 else 5
for j in range(10):
save_path = save_dir + "gesture/{}_gesture.gif".format(j)
if not os.path.exists(save_path):
x_train = x_gesture[j].detach().cpu().numpy()
frames = np.arange(kf_list[j])
plotUpperBody2D(x_train[:kf_list[j]], save_path, isRelative=isRelative, fps=fps, frames=frames)
save_path = save_dir + "gesture_allframe/{}_gesture.gif".format(j)
if not os.path.exists(save_path):
reconst = recon_g[j].detach().cpu().numpy()
frames = np.arange(opt.frames)
plotUpperBody2D(reconst, save_path, isRelative=isRelative, fps=fps, frames=frames)
save_path = save_dir + "gesture/{}_gesture_reconst.gif".format(j)
if not os.path.exists(save_path):
reconst = recon_g[j].detach().cpu().numpy()
frames = np.arange(kf_list[j])
plotUpperBody2D(reconst[:kf_list[j]], save_path, isRelative=isRelative, fps=fps, frames=frames)
# save_path = save_dir + "gesture_interp/{}_gesture_interp.gif".format(j)
# if not os.path.exists(save_path):
# frame_num = kf_list[j] * 12
# keyposes = recon_g[j].detach().cpu().numpy()[:kf_pred[j]]
# new_poses = interpolateKeypose(keyposes, frame=frame_num)
# plotUpperBody2D(new_poses, save_path, isRelative=isRelative, fps=25, frames= np.arange(frame_num))
# ############ Visualize Multi-model 2D Space ############
print("visualizing multimodal space")
text_f = text_feature.detach().cpu().numpy()
gest_f = gesture_feature.detach().cpu().numpy()
all = np.concatenate([gest_f, text_f])
# all_red = TSNE(n_components=2, random_state=0).fit_transform(all)
all_red = PCA(n_components=2, random_state=0).fit_transform(all)
gest_red = all_red[:opt.batch_size]
text_red = all_red[opt.batch_size:]
plt.figure()
for i in range(opt.batch_size):
# plt.text(gest_red[i][0], gest_red[i][1], str(i), color='darkorange')
# plt.text(text_red[i][0], text_red[i][1], str(i), color='blue')
plt.plot(gest_red[i][0], gest_red[i][1], marker="o", color='darkorange')
plt.plot(text_red[i][0], text_red[i][1], marker="o", color='blue')
for i in range(opt.batch_size):
for j in range(i+1, opt.batch_size):
if pos[i][j] == 1:
plt.plot([gest_red[i][0], gest_red[j][0]], [gest_red[i][1], gest_red[j][1]],
linestyle="dashed", color="r", marker="None", lw=1)
plt.xlim([-20, 20])
plt.ylim([-20, 20])
plt.show()
exit()
############ Visualize Attention Weight ############
a = attn.detach().cpu().numpy()
l = y_text.detach().cpu().numpy()
a_n = np.zeros([len(a[i]), len(a[i])])
for i in range(len(a)):
for j in range(len(a[i])):
a_n[i][j] = a[i][j] / sum(a[i])
for j in range(opt.batch_size):
tokens = tokenizer.convert_ids_to_tokens(x_text[j])
save_path = save_dir+"attention/{}.png".format(j)
visualizePrediction(tokens, a[j], l[j], save_path)
save_path = save_dir+"attention_norm/{}.png".format(j)
visualizePrediction(tokens, a_n[j], l[j], save_path)
exit()
g_features = np.array(g_features)
t_features = np.array(t_features)
kfs = np.array(kfs)
kf_preds = np.array(kf_preds)
print("Average of Keyframe Difference Pred - GT = ", np.mean(kf_preds - kfs))
plt.figure()
plt.hist([kf_preds, kfs], label=['Prediction', 'GT'])
plt.legend()
plt.show()
idx_list = np.where(pos[13] == 1)[0]
t = text_feature[idx_list].detach().cpu().numpy()
g = gesture_feature[idx_list].detach().cpu().numpy()
dm = np.zeros([len(t), len(g)])
for i in range(len(t)):
for j in range(len(g)):
dm[i][j] = np.linalg.norm(t[i] - g[j], ord=2)
for i in range(len(dm)):
print(np.argmin(dm[i]))
############ Output Pair Difference ############
diff = np.mean((g_features - t_features)**2)
print("Difference between text and gesture features: {}".format(diff))
diffs = np.mean((g_features - t_features)**2, axis=1)
plt.figure()
plt.hist(diffs, bins=10)
plt.xlim(0, 4)
plt.savefig(save_dir + "diff_{}.png".format(diff))
dataset = pickle.load(open(opt.dataset, "rb"))
gestures = dataset["gestures"]
if opt.frames > 100:
keyframes = [len(g) for g in dataset["gestures"]]
else:
keyframes = [len(g) for g in dataset["keyframe_list"]]
texts = dataset['remarks']
result = {'gestures': gestures, 'texts': texts, 'g_features': g_features, 't_features': t_features, 'keyframes': keyframes}
np.save("./data/features_keyframe_z={}d_g={}_distmat=vae_thresh={}.npy".format(opt.z_dim, opt.weight_g_cont, opt.distance_thresh), result)
print("Saved")
if __name__ == '__main__':
main(opt)