File size: 6,482 Bytes
15fa80a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import os 
import torch
import numpy as np
from torch.utils.tensorboard import SummaryWriter
import json
# import clip
from CLIP import clip

import options.option_transformer as option_trans
import models.vqvae as vqvae
import utils.utils_model as utils_model
import eval_trans_per as eval_trans
from dataset import dataset_TM_eval
import models.t2m_trans as trans
from options.get_eval_option import get_opt
from models.evaluator_wrapper import EvaluatorModelWrapper
import warnings
from tqdm import trange
warnings.filterwarnings('ignore')

##### ---- Exp dirs ---- #####
os.chdir('/root/autodl-tmp/SATO')
args = option_trans.get_args_parser()
torch.manual_seed(args.seed)

args.out_dir = os.path.join(args.out_dir, f'{args.exp_name}')
os.makedirs(args.out_dir, exist_ok = True)

##### ---- Logger ---- #####
logger = utils_model.get_logger(args.out_dir)
writer = SummaryWriter(args.out_dir)
logger.info(json.dumps(vars(args), indent=4, sort_keys=True))

from utils.word_vectorizer import WordVectorizer
w_vectorizer = WordVectorizer('./glove', 'our_vab')
val_loader = dataset_TM_eval.DATALoader(args.dataname, True, 32, w_vectorizer)

dataset_opt_path = 'checkpoints/kit/Comp_v6_KLD005/opt.txt' if args.dataname == 'kit' else 'checkpoints/t2m/Comp_v6_KLD005/opt.txt'

wrapper_opt = get_opt(dataset_opt_path, torch.device('cuda'))
eval_wrapper = EvaluatorModelWrapper(wrapper_opt)

##### ---- Network ---- #####

## load clip model and datasets

clip_model, clip_preprocess = clip.load(args.clip_path, device=torch.device('cuda'), jit=False)  # Must set jit=False for training
clip.model.convert_weights(clip_model)  # Actually this line is unnecessary since clip by default already on float16
clip_model.eval()
for p in clip_model.parameters():
    p.requires_grad = False

net = vqvae.HumanVQVAE(args, ## use args to define different parameters in different quantizers
                       args.nb_code,
                       args.code_dim,
                       args.output_emb_width,
                       args.down_t,
                       args.stride_t,
                       args.width,
                       args.depth,
                       args.dilation_growth_rate)


trans_encoder = trans.Text2Motion_Transformer(num_vq=args.nb_code, 
                                embed_dim=args.embed_dim_gpt, 
                                clip_dim=args.clip_dim, 
                                block_size=args.block_size, 
                                num_layers=args.num_layers, 
                                n_head=args.n_head_gpt, 
                                drop_out_rate=args.drop_out_rate, 
                                fc_rate=args.ff_rate)


print ('loading checkpoint from {}'.format(args.resume_pth))
ckpt = torch.load(args.resume_pth, map_location='cpu')
net.load_state_dict(ckpt['net'], strict=True)
net.eval()
net.cuda()

if args.resume_trans is not None:
    print ('loading transformer checkpoint from {}'.format(args.resume_trans))
    ckpt = torch.load(args.resume_trans, map_location='cpu')
    trans_encoder.load_state_dict(ckpt['trans'], strict=True)
trans_encoder.train()
trans_encoder.cuda()
print('checkpoints loading successfully')

fid = []
fid_per=[]
div = []
top1 = []
top2 = []
top3 = []
matching = []
multi = []
repeat_time = 20
fid_word_perb=[]
        
for i in range(repeat_time):
    print('repeat_time: ',i)
    best_fid,best_fid_word_perb,best_fid_per, best_iter, best_div, best_top1, best_top2, best_top3, best_matching, best_multi, writer, logger = eval_trans.evaluation_transformer_test(args.out_dir, val_loader, net, trans_encoder, logger, writer, 0, best_fid=1000,best_fid_word_perb=1000,best_fid_perturbation=1000, best_iter=0, best_div=100, best_top1=0, best_top2=0, best_top3=0, best_matching=100, best_multi=0, clip_model=clip_model, eval_wrapper=eval_wrapper, draw=False, savegif=False, save=False, savenpy=(i==0))
    fid.append(best_fid)
    fid_word_perb.append(best_fid_word_perb)
    fid_per.append(best_fid_per)
    div.append(best_div)
    top1.append(best_top1)
    top2.append(best_top2)
    top3.append(best_top3)
    matching.append(best_matching)
    multi.append(best_multi)

    # print('fid: ', sum(fid)/(i+1))
    # print('fid_per',sum(fid_per)/(i+1))
    # print('div: ', sum(div)/(i+1))
    # print('top1: ', sum(top1)/(i+1))
    # print('top2: ', sum(top2)/(i+1))
    # print('top3: ', sum(top3)/(i+1))
    # print('matching: ', sum(matching)/(i+1))
    # print('multi: ', sum(multi)/(i+1))


print('final result:')
print('fid: ', sum(fid)/repeat_time)
print('fid_word_perb',sum(fid_word_perb)/repeat_time)
print('fid_per',sum(fid_per)/repeat_time)
print('div: ', sum(div)/repeat_time)
print('top1: ', sum(top1)/repeat_time)
print('top2: ', sum(top2)/repeat_time)
print('top3: ', sum(top3)/repeat_time)
print('matching: ', sum(matching)/repeat_time)
# print('multi: ', sum(multi)/repeat_time)

fid = np.array(fid)
fid_word_perb=np.array(fid_word_perb)
fid_per=np.array(fid_per)
div = np.array(div)
top1 = np.array(top1)
top2 = np.array(top2)
top3 = np.array(top3)
matching = np.array(matching)
# multi = np.array(multi)
# msg_final = f"FID. {np.mean(fid):.3f}, FID_PERB.{np.mean(fid_per):.3f}conf. {np.std(fid)*1.96/np.sqrt(repeat_time):.3f}, Diversity. {np.mean(div):.3f}, conf. {np.std(div)*1.96/np.sqrt(repeat_time):.3f}, TOP1. {np.mean(top1):.3f}, conf. {np.std(top1)*1.96/np.sqrt(repeat_time):.3f}, TOP2. {np.mean(top2):.3f}, conf. {np.std(top2)*1.96/np.sqrt(repeat_time):.3f}, TOP3. {np.mean(top3):.3f}, conf. {np.std(top3)*1.96/np.sqrt(repeat_time):.3f}, Matching. {np.mean(matching):.3f}, conf. {np.std(matching)*1.96/np.sqrt(repeat_time):.3f}, Multi. {np.mean(multi):.3f}, conf. {np.std(multi)*1.96/np.sqrt(repeat_time):.3f}"
msg_final = f"FID. {np.mean(fid):.3f}, {np.std(fid)*1.96/np.sqrt(repeat_time):.3f}, FID_word_perb.{np.mean(fid_word_perb):.3f}, {np.std(fid_word_perb)*1.96/np.sqrt(repeat_time):.3f},FID_PERB.{np.mean(fid_per):.3f}, conf. {np.std(fid)*1.96/np.sqrt(repeat_time):.3f}, Diversity. {np.mean(div):.3f}, conf. {np.std(div)*1.96/np.sqrt(repeat_time):.3f}, TOP1. {np.mean(top1):.3f}, conf. {np.std(top1)*1.96/np.sqrt(repeat_time):.3f}, TOP2. {np.mean(top2):.3f}, conf. {np.std(top2)*1.96/np.sqrt(repeat_time):.3f}, TOP3. {np.mean(top3):.3f}, conf. {np.std(top3)*1.96/np.sqrt(repeat_time):.3f}, Matching. {np.mean(matching):.3f}, conf. {np.std(matching)*1.96/np.sqrt(repeat_time):.3f}, conf. {np.std(multi)*1.96/np.sqrt(repeat_time):.3f}"
logger.info(msg_final)