-
Notifications
You must be signed in to change notification settings - Fork 0
/
results_generation.py
243 lines (186 loc) · 10.2 KB
/
results_generation.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
import argparse
import os
import ruamel.yaml as yaml
import numpy as np
import random
import time
import datetime
import json
from pathlib import Path
import torch
import torch.nn as nn
import torch.backends.cudnn as cudnn
import sys
from models.model_vqa import ALBEF
from models.vit import interpolate_pos_embed
from models.tokenization_bert import BertTokenizer
import utils
from dataset import create_dataset, create_sampler, create_loader, vqa_collate_fn
from scheduler import create_scheduler
from optim import create_optimizer
from dataset.relation_dataset import Relation_val_dataset, pre_caption
import re
import glob
def main(args, config):
device = torch.device(args.device)
# fix the seed for reproducibility
seed = args.seed
torch.manual_seed(seed)
np.random.seed(seed)
random.seed(seed)
cudnn.benchmark = True
unus = ['[unused{}]'.format(x) for x in range(200,800)]
pos_token = ['@@']
pos_token.extend([f'[pos_{x}]' for x in range(args.position_res)])
pos_token.append('##')
postoken_dict = {}
tokenizer = BertTokenizer.from_pretrained('configs/vocab.txt')
for x,y in zip(unus, pos_token):
un_index = tokenizer.vocab[x]
tokenizer.vocab[y] = un_index
postoken_dict[y] = un_index
_ = tokenizer.vocab.pop(x)
tokenizer.basic_tokenizer.never_split.add(y)
postoken_dict.pop('@@')
postoken_dict.pop('##')
postoken_index = torch.randn(30522).bool()
postoken_index[:] = False
for x in postoken_dict.values():
postoken_index[x]=True
# data
dt1 = Relation_val_dataset(root = '', ann_file = config['test_file'], img_res=config['image_res'], replace = False)
samplers = [None]
col_fn = vqa_collate_fn
batch_size = args.batch_size
print('batch size ', batch_size)
data_loader = create_loader([dt1],samplers,
batch_size=[batch_size],
num_workers=[4],is_trains=[False],
collate_fns=[col_fn])[0]
# model
model = ALBEF(config=config, text_encoder='bert-base-uncased', text_decoder='bert-base-uncased', \
tokenizer=tokenizer, postoken_dict = postoken_dict)
print('Load checkpoint from ...', args.checkpoint)
checkpoint = torch.load(args.checkpoint, map_location='cpu')
state_dict = checkpoint['model']
# reshape positional embedding to accomodate for image resolution change
pos_embed_reshaped = interpolate_pos_embed(state_dict['visual_encoder.pos_embed'],model.visual_encoder)
state_dict['visual_encoder.pos_embed'] = pos_embed_reshaped
msg = model.load_state_dict(state_dict,strict=False)
print(msg)
postoken_dict_rev = {v:k for k,v in postoken_dict.items()}
postoken_dict_rev[int(tokenizer('@@').input_ids[-1])] = '@@'
postoken_dict_rev[int(tokenizer('##').input_ids[-1])] = '##'
model = model.cuda()
model.eval()
num_beams = args.num_beams
num_seq = args.num_seq
with_gt = args.with_gt
save_json = {}
start_point = args.chunk_size*args.chunk
end_point = args.chunk_size*(args.chunk+1)
print('chunk is ', args.chunk)
print('start from idx ', start_point)
print('end with idx ', end_point)
with torch.no_grad():
for i,(image, question, answer, weights_, n) in enumerate(data_loader):
if i < start_point:
continue
if i >= end_point:
break
if i % 50 == 0:
print(i)
json.dump(save_json, open('{}/{}_round{}_beam{}_numseq{}_{}_chunk{}.json'.format(args.result_dir, args.split, args.round, num_beams, num_seq, args.checkpoint[-6:-4], args.chunk),'w'))
image = image.cuda()
image_embeds = model.visual_encoder(image)
image_atts = torch.ones(image_embeds.size()[:-1],dtype=torch.long).to(image.device)
question_input = tokenizer(question, padding='longest', return_tensors="pt").to(image.device)
question_output = model.text_encoder(question_input.input_ids,
attention_mask = question_input.attention_mask,
encoder_hidden_states = image_embeds,
encoder_attention_mask = image_atts,
return_dict = True)
question_states = question_output.last_hidden_state.repeat_interleave(num_beams,dim=0)
question_atts = torch.ones(question_states.size()[:-1],dtype=torch.long).to(question_states.device)
model_kwargs = {"encoder_hidden_states": question_states, "encoder_attention_mask":question_atts}
bos_ids = torch.full((image.size(0),1),fill_value=101,device=image.device)
eos = int(tokenizer('@@').input_ids[-1])
outputs = model.text_decoder.generate(input_ids=bos_ids,
max_length=20,
min_length=1,
num_beams=num_beams,
num_return_sequences=num_seq,
eos_token_id=eos,
pad_token_id=model.tokenizer.pad_token_id,
return_dict_in_generate=True,
output_scores=True,
**model_kwargs)
sequences1 = outputs['sequences'] # batch*num_seq, 20
for ii in range(0,image.size(0)):
save_json[batch_size*i + ii] = {}
save_json[batch_size*i + ii]['round1_scores'] = outputs['sequences_scores'][ii*num_seq:(ii+1)*num_seq].tolist()
save_json[batch_size*i + ii]['round1_sequences'] = outputs['sequences'][ii*num_seq:(ii+1)*num_seq].tolist()
round2_scores = []
round2_raw_seqs = []
for seq_idx in range(0,len(sequences1)):
seq = sequences1[seq_idx]
image_embeds_seq = image_embeds[seq_idx//num_seq].unsqueeze(0)
image_atts_seq = image_atts[seq_idx//num_seq].unsqueeze(0)
seq1 = seq[seq.nonzero()].squeeze(1).unsqueeze(0).cuda()
num_beams_r2 = 5
question_states = question_output.last_hidden_state[seq_idx//num_seq].unsqueeze(0).repeat_interleave(num_beams_r2,dim=0)
question_atts = torch.ones(question_states.size()[:-1],dtype=torch.long).to(question_states.device)
model_kwargs = {"encoder_hidden_states": question_states, "encoder_attention_mask":question_atts}
max_len = 30
outputs = model.text_decoder.generate(input_ids=seq1,
max_length=max_len,
min_length=1,
num_beams=num_beams_r2,
num_return_sequences=1,
eos_token_id=model.tokenizer.sep_token_id,
pad_token_id=model.tokenizer.pad_token_id,
return_dict_in_generate=True,
output_scores=True,
**model_kwargs)
sequences2 = outputs['sequences']
round2_scores.append(outputs['sequences_scores'].tolist())
round2_raw_seqs.append(outputs['sequences'].tolist())
for ii in range(0,image.size(0)):
save_json[batch_size*i + ii]['round2_scores'] = round2_scores[ii*num_seq:(ii+1)*num_seq]
save_json[batch_size*i + ii]['round2_sequences'] = round2_raw_seqs[ii*num_seq:(ii+1)*num_seq]
json.dump(save_json, open('{}/{}_round{}_beam{}_numseq{}_{}_chunk{}.json'.format(args.result_dir, args.split, args.round, num_beams, num_seq, args.checkpoint[-6:-4], args.chunk),'w'))
return
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('--config', default='configs/relation_grounding.yaml')
parser.add_argument('--bert_config', default='configs/config_bert.json')
parser.add_argument('--root', default='checkpoints_folder/')
parser.add_argument('--output_dir', default='')
parser.add_argument('--text_encoder', default='bert-base-uncased')
parser.add_argument('--split', default='test')
parser.add_argument('--with_gt', action='store_true')
parser.add_argument('--device', default='cuda')
parser.add_argument('--position_res', default=512, type=int)
parser.add_argument('--seed', default=42, type=int)
parser.add_argument('--num_seq', default=50, type=int)
parser.add_argument('--num_beams', default=100, type=int)
parser.add_argument('--round', default=1, type=int)
parser.add_argument('--start', default=1, type=int)
parser.add_argument('--end', default=1, type=int)
parser.add_argument('--chunk', default=0, type=int) # 0,1,2,3
parser.add_argument('--chunk_size', default=5, type=int) # in each chunk, how many samples are processed
parser.add_argument('--batch_size', default=16, type=int)
args = parser.parse_args()
config = yaml.load(open(args.config, 'r'), Loader=yaml.Loader)
config['position_res'] = args.position_res
config['bert_config'] = args.bert_config
assert config['position_res'] == args.position_res
args.result_dir = os.path.join(args.root, args.output_dir, 'oidv6_results')
Path(args.result_dir).mkdir(parents=True, exist_ok=True)
yaml.dump(config, open(os.path.join(args.result_dir, 'config.yaml'), 'w'))
all_checkpoints = glob.glob(os.path.join(args.root, args.output_dir, '*.pth'))
all_checkpoints = all_checkpoints[args.start:args.end]
print(len(all_checkpoints))
for ckpt in all_checkpoints:
args.checkpoint = ckpt
main(args, config)