File size: 6,146 Bytes
012f8b5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import os
import torch
import torch.nn as nn
from torch.utils.data import Dataset, DataLoader
from torchvision import transforms
from transformers import ViTModel, BertTokenizerFast, BertConfig, BertLMHeadModel, AdamW
from PIL import Image, ImageFile
import pandas as pd
from tqdm import tqdm

# Increase the maximum image size limit to avoid DecompressionBombWarning
Image.MAX_IMAGE_PIXELS = None
# Allow loading truncated images
ImageFile.LOAD_TRUNCATED_IMAGES = True

# Check if CUDA is available
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
print(f"Using device: {device}")

# Define constants
VIT_MODEL_NAME = "google/vit-base-patch16-224"
BERT_MODEL_NAME = "dbmdz/bert-base-turkish-cased"  # Using a Turkish BERT model
model = "TeLVE_v1.0.pth"
MAX_LENGTH = 128
BATCH_SIZE = 8
EPOCHS = 5
LEARNING_RATE = 2e-5

class ImageCaptioningDataset(Dataset):
    def __init__(self, dataframe, img_dir, tokenizer):
        self.dataframe = dataframe
        self.img_dir = img_dir
        self.tokenizer = tokenizer
        self.transform = transforms.Compose([
            transforms.Resize((224, 224)),
            transforms.ToTensor(),
            transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
        ])

    def __len__(self):
        return len(self.dataframe)

    def __getitem__(self, idx):
        row = self.dataframe.iloc[idx]
        img_path = os.path.join(self.img_dir, row['photo_id'] + ".jpg")
        
        try:
            image = Image.open(img_path).convert('RGB')
            image = self.transform(image)
        except (FileNotFoundError, IOError):
            # Return None if the image is not found or cannot be opened
            return None
        
        caption = row['ai_description']

        # Check if caption is a valid string
        if not isinstance(caption, str):
            return None  # Skip the example if caption is not valid

        encoding = self.tokenizer(
            caption,
            add_special_tokens=True,
            max_length=MAX_LENGTH,
            padding='max_length',
            truncation=True,
            return_attention_mask=True,
            return_tensors='pt'
        )

        return {
            'pixel_values': image,
            'input_ids': encoding['input_ids'].squeeze(),
            'attention_mask': encoding['attention_mask'].squeeze(),
            'labels': encoding['input_ids'].squeeze()  # Use input_ids as labels for calculating loss
        }


class ImageCaptioningModel(nn.Module):
    def __init__(self, vit_model, bert_model):
        super(ImageCaptioningModel, self).__init__()
        self.vit = vit_model
        self.bert = bert_model
        self.linear = nn.Linear(self.vit.config.hidden_size, self.bert.config.hidden_size)

    def forward(self, pixel_values, input_ids, attention_mask, labels=None):
        image_features = self.vit(pixel_values).last_hidden_state
        image_features = self.linear(image_features)
        
        outputs = self.bert(input_ids=input_ids,
                            attention_mask=attention_mask,
                            encoder_hidden_states=image_features,
                            labels=labels,
                            return_dict=True)
        
        return outputs.loss, outputs.logits

def collate_fn(batch):
    # Filter out None values (skipped images)
    batch = list(filter(lambda x: x is not None, batch))
    if len(batch) == 0:
        return None
    return {key: torch.stack([item[key] for item in batch]) for key in batch[0]}

def train_vlm_model():
    # Load and preprocess the dataset
    encodings = ['utf-8', 'iso-8859-9', 'windows-1254']
    for encoding in encodings:
        try:
            df = pd.read_csv('./datasets/' + model + '.tsv000', sep='\t', encoding=encoding)
            print(f"Successfully read the file with {encoding} encoding.")
            break
        except UnicodeDecodeError:       
            print(f"Failed to read with {encoding} encoding. Trying next...")
    else:
        raise ValueError("Could not read the file with any of the specified encodings.")
    
    # Initialize the tokenizer
    tokenizer = BertTokenizerFast.from_pretrained(BERT_MODEL_NAME)

    # Create the dataset and dataloader
    dataset = ImageCaptioningDataset(df, '../download/images', tokenizer)
    dataloader = DataLoader(dataset, batch_size=BATCH_SIZE, shuffle=True, collate_fn=collate_fn)

    # Initialize the model components
    vit_model = ViTModel.from_pretrained(VIT_MODEL_NAME)
    bert_config = BertConfig.from_pretrained(BERT_MODEL_NAME)
    bert_config.is_decoder = True
    bert_config.add_cross_attention = True
    bert_model = BertLMHeadModel.from_pretrained(BERT_MODEL_NAME, config=bert_config)

    # Create the combined model
    model = ImageCaptioningModel(vit_model, bert_model)
    model.to(device)

    # Define optimizer
    optimizer = AdamW(model.parameters(), lr=LEARNING_RATE)

    # Training loop
    model.train()
    for epoch in range(EPOCHS):
        total_loss = 0
        progress_bar = tqdm(dataloader, desc=f"Epoch {epoch+1}/{EPOCHS}")
        for batch in progress_bar:
            if batch is None:
                continue
            
            pixel_values = batch['pixel_values'].to(device)
            input_ids = batch['input_ids'].to(device)
            attention_mask = batch['attention_mask'].to(device)
            labels = batch['labels'].to(device)

            optimizer.zero_grad()
            loss, _ = model(pixel_values, input_ids, attention_mask, labels)
            loss.backward()
            optimizer.step()

            total_loss += loss.item()
            progress_bar.set_postfix({'loss': loss.item()})

        print(f"Epoch {epoch+1}/{EPOCHS}, Average Loss: {total_loss/len(dataloader)}")

    # Save the model
    torch.save(model.state_dict(), "./models/" + model)
    tokenizer.save_pretrained("./tokenizer")

if __name__ == "__main__":
    train_vlm_model()