File size: 5,757 Bytes
a5f8a35
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import argparse
import json
import os
import random
from typing import Any, Dict, List

from loguru import logger
import torch
from torch.utils.data import DataLoader, DistributedSampler
from torch.nn.utils.rnn import pad_sequence
from tqdm import tqdm

import wordsegment as ws

from virtex.config import Config
from virtex.data import ZeroShotDataset

from virtex.data.tokenizers import SentencePieceBPETokenizer

from virtex.factories import TokenizerFactory, VisualBackboneFactory,TextualHeadFactory
from virtex.utils.checkpointing import CheckpointManager
from virtex.utils.common import common_parser
from virtex.utils.metrics import TopkAccuracy
import virtex.utils.distributed as dist


#importing classifier
from virtex.models.zero_shot_classification_eval import ZeroShotClassifier

ws.load()

# fmt: off
parser = common_parser(
    description="""Run image captioning inference on a pretrained model, and/or
    evaluate pretrained model on COCO Captions val2017 split."""
)
parser.add_argument(
    "--data-root", default=None,
    help="""Path to a directory containing image files to generate captions for imagenet.
    Default: COCO val2017 image directory as expected relative to project root."""
)
parser.add_argument(
    "--checkpoint-path", required=False,
    help="Path to load checkpoint and run captioning evaluation."
)
parser.add_argument(
    "--output", default=None,
    help="Path to save predictions as a JSON file."
)
parser.add_argument(
    "--calc-metrics", action="store_true",
    help="""Calculate CIDEr and SPICE metrics using ground truth COCO Captions.
    This flag should not be set when running inference on arbitrary images."""
)

parser.add_argument(
    "--idx_label_dict", default=None, required=False,
    help="""a dictionary that maps from lable index to label string for classification"""
)
parser.add_argument(
    "--is_redcaps", default=None, required=False,
    help="""a dictionary that maps from lable index to label string for"""
)
parser.add_argument(
    "--prompt_cls_sos", default=None, required=False,
    help="""a dictionary that maps from lable index to label string for"""
)
parser.add_argument(
    "--prompt_sos_eos", default=None, required=False,
    help="""a dictionary that maps from lable index to label string for"""
)
# fmt: on

print("###########")
print(os.getcwd() )
print("###########")

tokenizer = SentencePieceBPETokenizer("datasets_1/vocab/common_32k.model")

def main(_A: argparse.Namespace):
    if _A.num_gpus_per_machine == 0:
        # Set device as CPU if num_gpus_per_machine = 0.
        device = torch.device("cpu")
    else:
        # Get the current device (this will be zero here by default).
        device = torch.cuda.current_device()

    _C = Config(_A.config, _A.config_override)

    #tokenizer = TokenizerFactory.from_config(_C)
    
    if _A.data_root is None:
        _A.data_root = os.path.join(_C.DATA.ROOT, "val2017")
    
    if _A.is_redcaps == 1:
        model_dataset = 'redcaps'
    else:
        model_dataset = 'gcc or sbu'
        
    print(_A.idx_label_dict)
    
    val_dataset = ZeroShotDataset(data_root=_A.data_root,
                                  split="test/",
                                  label_map=_A.idx_label_dict,
                                  tokenizer=tokenizer,
                                  prompt_cls_sos=_A.prompt_cls_sos.replace("_", " "),
                                  prompt_sos_eos=_A.prompt_sos_eos.replace("_", " "))
    
    val_dataloader = DataLoader(
        val_dataset,
        batch_size= _C.OPTIM.BATCH_SIZE // dist.get_world_size(),
        num_workers=_A.cpu_workers,
        sampler=DistributedSampler(
            val_dataset,
            num_replicas=dist.get_world_size(),
            rank=dist.get_rank(),
        ),
        pin_memory=True,
        drop_last=False,
        collate_fn=val_dataset.collate_fn,
    )
    
    # Initialize model from a checkpoint
    visual = VisualBackboneFactory.from_config(_C)
    textual = TextualHeadFactory.from_config(_C)
    model = ZeroShotClassifier(visual,textual)
    ITERATION = CheckpointManager(model=model).load(_A.checkpoint_path)
    model.to(device).eval()
    
    ## setup distributed training 
    if dist.get_world_size() > 1:
        dist.synchronize()
        model = nn.parallel.DistributedDataParallel(
            model, device_ids=[device], find_unused_parameters=True
        )

    top_1 = TopkAccuracy(top_k=1)
    top_5 = TopkAccuracy(top_k=5)
    batch_num = 0
  
    
    for val_iteration, val_batch in tqdm(enumerate(val_dataloader, start=1)):
        val_batch["image"] = val_batch["image"].to(device)
        val_batch["caption_tokens"] = val_batch["caption_tokens"].to(device)
        val_batch["noitpac_tokens"] = val_batch["noitpac_tokens"] .to(device)
        val_batch["caption_lengths"] = val_batch["caption_lengths"].to(device)
        val_batch["label"] = val_batch["label"].to(device)        
         
        with torch.no_grad():
            classification_losses = model(val_batch)
            
        batch_num+=1
        top_1(classification_losses, val_batch["label"]) 
        top_1_acc = top_1.get_metric(reset=False)
        dist.average_across_processes(top_1_acc)

        top_5(classification_losses, val_batch["label"])
        top_5_acc = top_5.get_metric(reset=False)
        dist.average_across_processes(top_5_acc)

        logger.info(f"Iter: {val_iteration} | Top-1 accuracy: {top_1_acc} | Top-5 accuracy: {top_5_acc}")

   

if __name__ == "__main__":
    _A = parser.parse_args()
    #if _A.num_gpus_per_machine > 1:
    #    raise ValueError("Using multiple GPUs is not supported for this script.")

    # No distributed training here, just a single process.
    main(_A)