/
run.py
169 lines (141 loc) · 8.86 KB
/
run.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
import argparse
import logging
import sys
sys.path.append("..")
import torch
import numpy as np
import random
from torchvision import transforms
from torch.utils.data import DataLoader
from cores.gene.model import MRE
from transformers import CLIPProcessor, CLIPModel
from transformers import CLIPConfig
from processor.dataset import MMREProcessor, NewMMREDatasetForIB
from cores.gene.model import Trainer
import os
os.environ["CUDA_VISIBLE_DEVICES"] = "0"
import warnings
warnings.filterwarnings("ignore", category=UserWarning)
from torch.utils.tensorboard import SummaryWriter
logging.basicConfig(format = '%(asctime)s - %(levelname)s - %(name)s - %(message)s',
datefmt = '%m/%d/%Y %H:%M:%S',
level = logging.INFO)
logger = logging.getLogger(__name__)
MODEL_CLASS = {
'bert': (MMREProcessor, NewMMREDatasetForIB),
}
DATA_PATH = {
'MRE': {'train': 'data/vsg_tsg/ours_train.json',
'dev': 'data/vsg_tsg/ours_valid.json',
'test': 'data/vsg_tsg/ours_test.json',
'vbow': 'data/vsg_tsg/vbow.pk',
'tbow': 'data/vsg_tsg/tbow.pk'
}
}
IMG_PATH = {
'MRE': {'train': 'data/img_org/train/',
'dev': 'data/img_org/val/',
'test': 'data/img_org/test'}}
def set_seed(seed=2021):
"""set random seed"""
torch.manual_seed(seed)
torch.cuda.manual_seed_all(seed)
torch.backends.cudnn.deterministic = True
torch.backends.cudnn.enabled = True
torch.backends.cudnn.benchmark = True
np.random.seed(seed)
random.seed(seed)
def main():
parser = argparse.ArgumentParser()
parser.add_argument('--pretrain_name', default='openai/clip-vit-base-patch32', type=str, help="The name of pretrained model")
parser.add_argument('--dataset_name', default='MRE', type=str, help="The name of example_dataset.")
parser.add_argument('--num_epochs', default=30, type=int, help="Training epochs")
parser.add_argument('--device', default='cuda', type=str, help="cuda or cpu")
parser.add_argument('--batch_size', default=32, type=int, help="batch size")
parser.add_argument('--lr_pretrained', default=2e-5, type=float, help="pre-trained learning rate")
parser.add_argument('--lr_main', default=2e-4, type=float, help="learning rate")
parser.add_argument('--warmup_ratio', default=0.01, type=float)
parser.add_argument('--eval_begin_epoch', default=1, type=int)
parser.add_argument('--seed', default=1234, type=int, help="random seed, default is 1")
parser.add_argument('--load_path', default=None, type=str, help="Load model from load_path")
parser.add_argument('--save_path', default='ckpt', type=str, help="save model at save_path")
parser.add_argument('--write_path', default=None, type=str, help="do_test=True, predictions will be write in write_path")
parser.add_argument('--notes', default="", type=str, help="input some remarks for making save path dir.")
parser.add_argument('--do_train', action='store_true') # , action='store_true'
parser.add_argument('--do_test', action='store_true')
parser.add_argument('--do_predict', action='store_true')
parser.add_argument('--max_seq', default=40, type=int)
parser.add_argument('--max_obj', default=40, type=int)
parser.add_argument('--hid_size', default=768, type=int, help="hidden state size")
parser.add_argument('--num_layers', default=2, type=int, help="number of refine layers")
parser.add_argument('--beta', default=0.01, type=float, help="Default is 1e-2")
parser.add_argument("--num_per", type=int, default=16, help="Default is 16")
parser.add_argument("--feature_denoise", type=bool, default=True, help="Default is False.")
parser.add_argument("--top_k", type=int, default=10, help="Default is 10.")
parser.add_argument("--epsilon", type=float, default=0.3, help="Default is 0.3.")
parser.add_argument("--temperature", type=float, default=0.1, help="Default is 0.1.")
parser.add_argument("--graph_skip_conn", type=float, default=0.0, help="Default is 0.0.")
parser.add_argument("--graph_include_self", type=bool, default=True, help="Default is True.")
parser.add_argument("--dropout", type=float, default=0.5, help="Default is 0.0")
parser.add_argument("--graph_type", type=str, default="epsilonNN", help="epsilonNN, KNN, prob")
parser.add_argument("--graph_metric_type", type=str, default="multi_mlp")
parser.add_argument("--repar", type=bool, default=True, help="Default is True.")
parser.add_argument("--threshold", type=float, default=0.2, help="Default is 0.2.")
parser.add_argument("--prior_mode", type=str, default="Gaussian", help="Default is Gaussian.")
parser.add_argument("--is_IB", type=bool, default=True, help="Default is True.")
parser.add_argument("--eta1", type=float, default=0.7, help="Default is 0.7")
parser.add_argument("--eta1", type=float, default=0.9, help="Default is 0.9")
parser.add_argument("--text_bow_size", type=int, default=2000, help="Default is 2000")
parser.add_argument("--visual_bow_size", type=int, default=2000, help="Default is 2000")
parser.add_argument("--neighbor_num", type=int, default=2, help="Default is 2")
parser.add_argument("--topic_keywords_number", type=int, default=10, help="Default is 10")
parser.add_argument("--topic_number", type=int, default=10, help="Default is 10")
args = parser.parse_args()
data_path, img_path, aux_path = DATA_PATH[args.dataset_name], IMG_PATH[args.dataset_name], AUX_PATH[args.dataset_name]
data_process, dataset_class = MODEL_CLASS[args.model_name]
re_path = 'data/ours_rel2id.json'
transform = transforms.Compose([
transforms.Resize(256),
transforms.CenterCrop(224),
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])])
set_seed(args.seed) # set seed, default is 1
if args.save_path is not None: # make save_path dir
args.save_path = os.path.join(args.save_path, args.model_name, args.dataset_name+"_"+str(args.batch_size)+"_"+str(args.lr)+"_"+args.notes)
if not os.path.exists(args.save_path):
os.makedirs(args.save_path, exist_ok=True)
print(args)
logdir = "logs/" + args.model_name + "_"+args.dataset_name + "_"+str(args.batch_size) + "_" + str(args.lr) + args.notes
writer = SummaryWriter(log_dir=logdir)
if args.do_train:
clip_vit, clip_processor, aux_processor, rcnn_processor = None, None, None, None
clip_processor = CLIPProcessor.from_pretrained(args.vit_name)
aux_processor = CLIPProcessor.from_pretrained(args.vit_name)
aux_processor.feature_extractor.size, aux_processor.feature_extractor.crop_size = args.aux_size, args.aux_size
rcnn_processor = CLIPProcessor.from_pretrained(args.vit_name)
rcnn_processor.feature_extractor.size, rcnn_processor.feature_extractor.crop_size = args.rcnn_size, args.rcnn_size
clip_model = CLIPModel.from_pretrained(args.vit_name)
clip_vit = clip_model.vision_model
clip_text = clip_model.text_model
processor = data_process(data_path, re_path, args.bert_name, args.vit_name, clip_processor=clip_processor, aux_processor=aux_processor, rcnn_processor=rcnn_processor)
train_dataset = dataset_class(processor, transform, img_path, aux_path, args.max_seq, aux_size=args.aux_size, rcnn_size=args.rcnn_size, mode='train', max_obj_num=args.max_obj_num)
train_dataloader = DataLoader(train_dataset, batch_size=args.batch_size, shuffle=True, num_workers=4, pin_memory=True)
dev_dataset = dataset_class(processor, transform, img_path, aux_path, args.max_seq, aux_size=args.aux_size, rcnn_size=args.rcnn_size, mode='dev', max_obj_num=args.max_obj_num)
dev_dataloader = DataLoader(dev_dataset, batch_size=args.batch_size, shuffle=False, num_workers=4, pin_memory=True)
test_dataset = dataset_class(processor, transform, img_path, aux_path, args.max_seq, aux_size=args.aux_size, rcnn_size=args.rcnn_size, mode='test', max_obj_num=args.max_obj_num)
test_dataloader = DataLoader(test_dataset, batch_size=args.batch_size, shuffle=False, num_workers=4, pin_memory=True)
re_dict = processor.get_relation_dict()
num_labels = len(re_dict)
tokenizer = processor.tokenizer
# test
vision_config = CLIPConfig.from_pretrained(args.vit_name).vision_config
text_config = CLIPConfig.from_pretrained(args.vit_name).text_config
model = MRE(args, vision_config, text_config, clip_vit, clip_text, num_labels,
args.text_bow_size, args.visual_bow_size, tokenizer, processor)
trainer = Trainer(train_data=train_dataloader, dev_data=dev_dataloader, test_data=test_dataloader, re_dict=re_dict, model=model, args=args, logger=logger, writer=writer)
trainer.train()
torch.cuda.empty_cache()
writer.close()
if __name__ == "__main__":
main()