-
Notifications
You must be signed in to change notification settings - Fork 3
/
eval.py
166 lines (141 loc) · 6.95 KB
/
eval.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
import argparse
import copy
import json
import os
import numpy as np
import torch
import torch.utils.data as data
from thop import profile
from datasets.PianoPose import PianoPose
from models.piano2posi import Piano2Posi
from models.denoise_diffusion import GaussianDiffusion1D_piano2pose, Unet1D
from models.evaluate import fid, loc_distance, MW2, acceleration, EmbeddingSpaceEvaluator
DEBUG = 0
def get_args():
parser = argparse.ArgumentParser()
parser.add_argument("--exp_path", type=str, default="logs/diffusion_posiguide")
# dataset
parser.add_argument("--data_root", type=str, default='/hdd/data3/piano-bilibili')
parser.add_argument("--mode", type=str, default='test')
parser.add_argument("--valid_batch_size", type=int, default=1)
args = parser.parse_args()
return args
def main():
# get args
args_exp = get_args()
args = copy.copy(args_exp)
with open(args.exp_path + '/args.txt', 'r') as f:
args.__dict__ = json.load(f)
args.preload = False
args.data_root = args_exp.data_root
args.up_list = ['1467634', '66685747']
# get dataloader
valid_loader = data.DataLoader(
PianoPose(args=args, phase=args_exp.mode),
batch_size=args_exp.valid_batch_size)
# load model
args_piano2posi = copy.copy(args)
with open(args_exp.exp_path + '/args_posi.txt', 'r') as f:
dic = json.load(f)
dic['hidden_type'] = args.hidden_type if 'hidden_type' in args.__dict__.keys() else 'audio_f'
args_piano2posi.__dict__ = dic
piano2posi = Piano2Posi(args_piano2posi)
if 'hidden_type' in args.__dict__.keys():
if args.hidden_type == 'audio_f':
cond_dim = 768 if 'base' in args_piano2posi.wav2vec_path else 1024
elif args.hidden_type == 'hidden_f':
cond_dim = args_piano2posi.feature_dim
elif args.hidden_type == 'both':
cond_dim = args_piano2posi.feature_dim + (768 if 'base' in args_piano2posi.wav2vec_path else 1024)
else:
cond_dim = 768 if 'base' in args_piano2posi.wav2vec_path else 1024
unet = Unet1D(
dim=args.unet_dim,
dim_mults=(1, 2, 4, 8),
channels=args.bs_dim,
remap_noise=args.remap_noise if 'remap_noise' in args.__dict__ else True,
condition=True,
guide=args.xyz_guide,
guide_dim=6 if args.xyz_guide else 0,
condition_dim=cond_dim,
encoder_type=args.encoder_type if 'encoder_type' in args.__dict__ else 'none',
num_layer=args.num_layer if 'num_layer' in args.__dict__ else None
)
timesteps = args.timesteps
model = GaussianDiffusion1D_piano2pose(
unet,
piano2posi,
seq_length=args.train_sec * 30,
timesteps=timesteps,
objective='pred_v',
)
model_path = sorted([i for i in os.listdir(args_exp.exp_path) if 'ckpt' in i], key=lambda x: int(x.split('-')[1].split('=')[1]))[-1]
print('load from:', model_path)
state_dict = torch.load(args_exp.exp_path + '/' + model_path, map_location='cpu')['state_dict']
model.load_state_dict(state_dict)
model.to('cuda')
fid2 = EmbeddingSpaceEvaluator(embed_net_path='checkpoints/gesture_autoencoder_checkpoint_best.bin')
scale = torch.tensor([1.5, 1.5, 25]).cuda()
metrics = {'right': {'fid': [], 'MW': [], 'dist':[], 'smooth': []},
'left': {'fid': [], 'MW': [], 'dist':[], 'smooth': []}}
results = {'right_pose': [], 'right_trans': [], 'left_pose': [], 'left_trans': []}
gts = {'right_pose': [], 'right_trans': [], 'left_pose': [], 'left_trans': []}
with torch.no_grad():
for v_idx, batch in enumerate(valid_loader):
for key in batch.keys():
batch[key] = batch[key].cuda()
audio, right_pose, left_pose = batch['audio'], batch['right'], batch['left']
frame_num = left_pose.shape[1]
#generation
pose_hat, guide = model.sample(audio, frame_num, right_pose.shape[0])
pose_hat = pose_hat.permute(0, 2, 1)
guide = guide.permute(0, 2, 1)
# re-normalize
right_pose_pred = (pose_hat[:, :, :args.bs_dim // 2] * np.pi).cpu().numpy()
left_pose_pred = (pose_hat[:, :, args.bs_dim // 2:] * np.pi).cpu().numpy()
right_trans_pred = (guide[:, :, :3] * scale).cpu().numpy()
left_trans_pred = (guide[:, :, 3:] * scale).cpu().numpy()
right_trans = right_pose[:, :, 1:4].cpu().numpy()
left_trans = left_pose[:, :, 1:4].cpu().numpy()
right_pose = right_pose[:, :, 4:].cpu().numpy()
left_pose = left_pose[:, :, 4:].cpu().numpy()
results['right_pose'].append(right_pose_pred)
results['right_trans'].append(right_trans_pred)
results['left_pose'].append(left_pose_pred)
results['left_trans'].append(left_trans_pred)
gts['right_pose'].append(right_pose)
gts['right_trans'].append(right_trans)
gts['left_pose'].append(left_pose)
gts['left_trans'].append(left_trans)
for i in range(right_pose.shape[0]):
right_fid = fid(right_pose[i], right_pose_pred[i])
left_fid = fid(left_pose[i], left_pose_pred[i])
right_loc_dist = loc_distance(right_trans[i, :, :2], right_trans_pred[i, :, :2])
left_loc_dist = loc_distance(left_trans[i, :, :2], left_trans_pred[i, :, :2])
right_MW2 = MW2(right_pose[i], right_pose_pred[i])
left_MW2 = MW2(left_pose[i], left_pose_pred[i])
right_smooth = np.mean(np.abs(acceleration(right_pose_pred[i]) - acceleration(right_pose[i])))
left_smooth = np.mean(np.abs(acceleration(left_pose_pred[i]) - acceleration(left_pose[i])))
metrics['right']['fid'].append(right_fid)
metrics['left']['fid'].append(left_fid)
metrics['right']['dist'].append(right_loc_dist)
metrics['left']['dist'].append(left_loc_dist)
metrics['right']['MW'].append(right_MW2)
metrics['left']['MW'].append(left_MW2)
metrics['right']['smooth'].append(right_smooth)
metrics['left']['smooth'].append(left_smooth)
fid2.push_samples(np.concatenate([right_pose_pred, left_pose_pred], 2),
np.concatenate([right_pose, left_pose], 2))
for key in metrics['right'].keys():
metrics['right'][key] = np.mean(metrics['right'][key])
metrics['left'][key] = np.mean(metrics['left'][key])
metrics['FID2'], metrics['F_dist'] = fid2.get_scores()
print(metrics)
audio, right_pose, left_pose = batch['audio'], batch['right'], batch['left']
gt = torch.cat([right_pose[:, :, 4:], left_pose[:, :, 4:]], 2).permute(0, 2, 1) / np.pi
# PARAMs
macs, params = profile(model, inputs=(gt[0:1].cuda(), audio[0:1].cuda()))
print(f"macs = {macs / 1e9}G")
print(f"params = {params / 1e6}M")
if __name__ == "__main__":
main()