-
Notifications
You must be signed in to change notification settings - Fork 11
/
process_dataset.py
191 lines (177 loc) · 9.05 KB
/
process_dataset.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
import os
import src.Datasets.BaseLoader as mBaseLoader
from src.Datasets.BatchProcessor import BatchRotateYCenterXZ
import torch
import numpy as np
import random
from src.Datasets.Style100Processor import StyleLoader,Swap100StyJoints,bvh_to_binary,save_skeleton
import src.utils.BVH_mod as BVH
from src.utils.motion_process import subsample
from src.Datasets.BaseLoader import BasedDataProcessor,BasedLoader,DataSetType,MotionDataLoader,WindowBasedLoader
class TransitionProcessor(BasedDataProcessor):
def __init__(self,ref_id):
super(TransitionProcessor, self).__init__()
self.process = BatchRotateYCenterXZ()
self.ref_id = ref_id
def __call__(self, dict, skeleton,motionDataLoader,ratio=1.0):
offsets, hip_pos, quats = dict["offsets"], dict["hip_pos"], dict["quats"]
stat = self._calculate_stat(offsets,hip_pos,quats,skeleton,ratio)
return {"offsets":offsets,"hip_pos":hip_pos,"quats":quats,**stat}
def _concat(self,local_quat,offsets,hip_pos,ratio=0.2):
if(ratio>=1.0):
local_quat = torch.from_numpy(np.concatenate(local_quat, axis=0))
offsets = torch.from_numpy(np.concatenate((offsets), axis=0))
hip_pos = torch.from_numpy(np.concatenate((hip_pos), axis=0))
else:
length = int(len(local_quat)*ratio)+1
idx = []
for i in range(length):
idx.append(random.randrange(0,len(local_quat)))
sample = lambda x:[x[i] for i in idx]
local_quat = torch.from_numpy(np.concatenate(sample(local_quat), axis=0))
offsets = torch.from_numpy(np.concatenate(sample(offsets), axis=0))
hip_pos =torch.from_numpy(np.concatenate(sample(hip_pos), axis=0))
return local_quat,offsets,hip_pos
def calculate_pos_statistic(self, pos):
'''pos:N,T,J,3'''
mean = np.mean(pos, axis=(0, 1))
std = np.std(pos, axis=(0, 1))
std = np.mean(std)
return mean, std
def calculate_rotation_mean(self,rotation):
mean = np.mean(rotation,axis=(0,1))
std = np.std(rotation,axis=(0,1))
std = np.mean(std)
return mean ,std
def calculate_statistic(self, local_pos,local_rot):
pos_mean, pos_std = self.calculate_pos_statistic(local_pos[:,:,1:,:].cpu().numpy())
vel_mean, vel_std = self.calculate_pos_statistic((local_pos[:, 1:, 1:, :] - local_pos[:, :-1, 1:, :]).cpu().numpy())
hipv_mean, hipv_std = self.calculate_pos_statistic((local_pos[:, 1:, 0:1, :] - local_pos[:, :-1, 0:1, :]).cpu().numpy())
rot_mean,rot_std = self.calculate_rotation_mean(local_rot[:,:,1:,:].cpu().numpy())
rotv_mean,rotv_std = self.calculate_rotation_mean((local_rot[:,1:,1:,:]-local_rot[:,:-1,1:,:]).cpu().numpy())
hipr_mean, hipr_std = self.calculate_rotation_mean(local_rot[:, :, 0:1, :].cpu().numpy())
hiprv_mean, hiprv_std = self.calculate_rotation_mean((local_rot[:, 1:, 0:1, :] - local_rot[:, :-1, 0:1, :]).cpu().numpy())
return {"pos_stat": [pos_mean, pos_std], "rot_stat":[rot_mean,rot_std],"vel_stat":[vel_mean,vel_std],
"rotv_stat":[rotv_mean,rotv_std],"hipv_stat":[hipv_mean,hipv_std],"hipr_stat":[hipr_mean,hipr_std],"hiprv_stat":[hiprv_mean,hiprv_std]}
def _calculate_stat(self,offsets,hip_pos,local_quat,skeleton,ratio):
local_quat, offsets, hip_pos = self._concat(local_quat,offsets,hip_pos,ratio)
global_positions, global_rotations = skeleton.forward_kinematics(local_quat, offsets, hip_pos)
local_pos,local_rot = self.process(global_positions,local_quat, self.ref_id)
return self.calculate_statistic(local_pos,local_rot)
def read_style_bvh(style,content,clip=None):
swap_joints = Swap100StyJoints()
anim = BVH.read_bvh(os.path.join("MotionData/100STYLE/",style,style+"_"+content+".bvh"),remove_joints=swap_joints)
if (clip != None):
anim.quats = anim.quats[clip[0]:clip[1], ...]
anim.hip_pos = anim.hip_pos[clip[0]:clip[1], ...]
anim = subsample(anim,ratio=2)
return anim
def processStyle100Benchmark( window, overlap):
style_loader = StyleLoader()
processor = None
bloader = mBaseLoader.WindowBasedLoader(window=window, overlap=overlap, subsample=1)
style_loader.setup(bloader, processor)
style_loader.load_dataset("+phase_gv10")
def split_window(motions):
for style in motions.keys():
styles = []
# print(style)
if len(motions[style].keys()):
dict = motions[style].copy()
for content in motions[style].keys():
motions[style][content] = bloader.append_dicts(motions[style][content])
for content in dict.keys():
if dict[content]['hip_pos'][0].shape[0]>=120:
o = dict[content]['offsets'][0]
h = dict[content]['hip_pos'][0][0:120]
q = dict[content]['quats'][0][0:120]
styles.append({"offsets": o, "hip_pos": h, "quats": q})
motions[style]['style'] = styles
result = {}
for style_name in motions.keys():
# print(motions.keys())
o, h, q, a, s, b, f = [], [], [], [], [], [], []
for content_name in motions[style_name]:
if content_name == 'style':
continue
dict = motions[style_name][content_name]
o += dict['offsets']
h += dict['hip_pos']
q += dict['quats']
a += dict['A']
s += dict['S']
b += dict['B']
f += dict['F']
# i += 1
style = motions[style_name]['style']
motion = {"offsets": o, "hip_pos": h, "quats": q, "A": a, "S": s, "B": b, "F": f}
result[style_name] = {"motion":motion, "style":style}
return result
style_loader.test_dict = split_window(style_loader.test_motions)
style_loader.save_to_binary("style100_benchmark_65_25", style_loader.test_dict)
def processTransitionPhaseDatasetForStyle100(window,overlap):
style_loader = StyleLoader()
window_loader = mBaseLoader.WindowBasedLoader(window, overlap, 1)
processor = None #MotionPuzzleProcessor()
style_loader.setup(window_loader, processor)
style_loader.load_dataset("+phase_gv10")
def split_window(motions):
#motions = style_loader.all_motions
for style in motions.keys():
for content in motions[style].keys():
motions[style][content] = window_loader.append_dicts(motions[style][content])
return motions
style_loader.train_motions = split_window(style_loader.train_motions)
style_loader.test_motions = split_window(style_loader.test_motions)
#style_loader.save_part_to_binary("motionpuzzle_statistics", ["pos_stat", "vel_stat", "rot_stat"])
style_loader.save_dataset("+phase_gv10" + window_loader.get_postfix_str())
print()
def processDeepPhaseForStyle100(window,overlap):
from src.Datasets.DeepPhaseDataModule import DeepPhaseProcessor
style_loader = StyleLoader()
window_loader = mBaseLoader.WindowBasedLoader(window,overlap,1)
processor = DeepPhaseProcessor(1./30)
style_loader.setup(window_loader,processor)
style_loader.process_from_binary()
style_loader.save_train_test_dataset("deep_phase_gv")
def splitStyle100TrainTestSet():
style_loader = StyleLoader()
print("Divide the data set to train set and test set")
style_loader.split_from_binary()
print("argument datasets")
style_loader.augment_dataset()
print("down")
if __name__ == '__main__':
from argparse import ArgumentParser
parser = ArgumentParser()
parser.add_argument("--preprocess", action="store_true")
parser.add_argument("--train_phase_model", action="store_true")
parser.add_argument("--add_phase_to_dataset", action="store_true")
parser.add_argument("--model_path",type=str,default="./results/deephase_sty/myResults/31/epoch=161-step=383778-v1.ckpt")
parser.add_argument("--train_manifold_model", action="store_true")
parser.add_argument("--train_sampler_model", action="store_true")
parser.add_argument("--benchmarks", action="store_true")
args = parser.parse_args()
if(args.preprocess==True):
print("######## convert all bvh files to binary files################")
bvh_to_binary()
save_skeleton()
print("\nConvert down\n")
print("Divide the dataset to train set and test set, and then argument datasets.")
splitStyle100TrainTestSet()
elif(args.train_phase_model==True):
processDeepPhaseForStyle100(62,2)
elif(args.add_phase_to_dataset==True):
from add_phase_to_dataset import add_phase_to_100Style
style100_info = {
"model_path":args.model_path,
"dt": 1. / 30,
"window": 61
}
add_phase_to_100Style(style100_info)
elif(args.train_manifold_model==True):
processTransitionPhaseDatasetForStyle100(61,21)
elif(args.train_sampler_model==True):
processTransitionPhaseDatasetForStyle100(120,0)
elif(args.benchmarks==True):
processStyle100Benchmark(65,25)