File size: 6,371 Bytes
114e0ca
 
 
 
 
30dbc69
4d0c7ca
30dbc69
4d0c7ca
 
30dbc69
4d0c7ca
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
30dbc69
4d0c7ca
30dbc69
4d0c7ca
 
 
 
 
30dbc69
4d0c7ca
 
 
30dbc69
 
 
 
114e0ca
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
30dbc69
114e0ca
 
 
 
 
 
 
 
 
 
 
4d0c7ca
 
 
 
 
 
 
 
114e0ca
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c8b4cd2
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
import torch
import torch.nn as nn
from torchvision import models
import pickle
from pathlib import Path
import sys
import logging

# Configure logger
logger = logging.getLogger(__name__)

class Vocabulary:    
    def __init__(self, freq_threshold=5):
        self.freq_threshold = freq_threshold
        self.word2idx = {}
        self.idx2word = {}
        self.idx = 0
        
        # Special tokens
        self.pad_token = "<PAD>"
        self.start_token = "<SOS>"
        self.end_token = "<EOS>"
        self.unk_token = "<UNK>"
        
        # Add special tokens
        for token in [self.pad_token, self.start_token, self.end_token, self.unk_token]:
            self.add_word(token)
    
    def add_word(self, word):
        """Add a word to the vocabulary"""
        if word not in self.word2idx:
            self.word2idx[word] = self.idx
            self.idx2word[self.idx] = word
            self.idx += 1
    
    def __len__(self):
        return len(self.word2idx)
    
    def __call__(self, word):
        """Convert word to index"""
        if word not in self.word2idx:
            return self.word2idx[self.unk_token]
        return self.word2idx[word]
    
    def decode(self, indices):
        """Convert indices back to words"""
        return [self.idx2word[idx] for idx in indices if idx in self.idx2word]

import __main__
setattr(__main__, "Vocabulary", Vocabulary)


class EncoderCNN(nn.Module):    
    def __init__(self, embed_size):
        super(EncoderCNN, self).__init__()
        resnet = models.resnet50(pretrained=False)
        modules = list(resnet.children())[:-1]
        self.resnet = nn.Sequential(*modules)
        self.fc = nn.Linear(resnet.fc.in_features, embed_size)
        self.bn = nn.BatchNorm1d(embed_size, momentum=0.01)
    
    def forward(self, images):
        features = self.resnet(images)
        features = features.view(features.size(0), -1)
        features = self.fc(features)
        features = self.bn(features)
        return features

class DecoderLSTM(nn.Module):    
    def __init__(self, embed_size, hidden_size, vocab_size, num_layers, dropout=0.5):
        super(DecoderLSTM, self).__init__()
        self.embed = nn.Embedding(vocab_size, embed_size)
        self.lstm = nn.LSTM(embed_size, hidden_size, num_layers, 
                           batch_first=True, dropout=dropout if num_layers > 1 else 0)
        self.dropout = nn.Dropout(dropout)
        self.fc = nn.Linear(hidden_size, vocab_size)
    
    def forward(self, features, captions):
        embeddings = self.embed(captions)
        embeddings = torch.cat((features.unsqueeze(1), embeddings), dim=1)
        hiddens, _ = self.lstm(embeddings)
        outputs = self.fc(hiddens)
        return outputs
    
    def sample(self, features, max_length=50):
        batch_size = features.size(0)
        captions = []
        states = None
        inputs = features.unsqueeze(1)
        
        for _ in range(max_length):
            hiddens, states = self.lstm(inputs, states)
            outputs = self.fc(hiddens.squeeze(1))
            predicted = outputs.argmax(dim=1)
            captions.append(predicted)
            inputs = self.embed(predicted).unsqueeze(1)
        
        captions = torch.stack(captions, dim=1)
        return captions

class ImageCaptioningModel(nn.Module):    
    def __init__(self, embed_size, hidden_size, vocab_size, num_layers, dropout=0.5):
        super(ImageCaptioningModel, self).__init__()
        self.encoder = EncoderCNN(embed_size)
        self.decoder = DecoderLSTM(embed_size, hidden_size, vocab_size, num_layers, dropout)
    
    def forward(self, images, captions):
        features = self.encoder(images)
        outputs = self.decoder(features, captions)
        return outputs
    
    def generate_caption(self, images, max_length=50):
        features = self.encoder(images)
        captions = self.decoder.sample(features, max_length)
        return captions

class ActionRecognitionModel(nn.Module):    
    def __init__(self, num_classes, dropout=0.5):
        super(ActionRecognitionModel, self).__init__()
        self.backbone = models.resnet50(pretrained=False)
        num_features = self.backbone.fc.in_features
        
        self.backbone.fc = nn.Sequential(
            nn.Dropout(dropout),
            nn.Linear(num_features, 512),
            nn.ReLU(),
            nn.BatchNorm1d(512),
            nn.Dropout(dropout),
            nn.Linear(512, num_classes)
        )
    
    def forward(self, x):
        return self.backbone(x)


def load_caption_model(device, model_dir=None):
    if model_dir is None:
        model_dir = Path(__file__).parent / 'models'
    else:
        model_dir = Path(model_dir)
    
    # Load configuration
    with open(model_dir / 'caption_model_config.pkl', 'rb') as f:
        config = pickle.load(f)
    
    # Load vocabulary
    try:
        with open(model_dir / 'vocab.pkl', 'rb') as f:
            vocab = pickle.load(f)
        logger.info(f"Vocabulary loaded successfully. Size: {len(vocab)}")
    except Exception as e:
        logger.error(f"Failed to load vocabulary: {e}")
        raise e
            
    # Create model
    model = ImageCaptioningModel(
        embed_size=config['embed_size'],
        hidden_size=config['hidden_size'],
        vocab_size=config['vocab_size'],
        num_layers=config['num_layers'],
        dropout=config['dropout']
    )
    
    # Load weights
    model.load_state_dict(torch.load(model_dir / 'caption_model_final.pth', 
                                     map_location=device))
    model = model.to(device)
    model.eval()
    
    return model, vocab

def load_action_model(device, model_dir=None):
    """Load action recognition model"""
    if model_dir is None:
        model_dir = Path(__file__).parent / 'models'
    else:
        model_dir = Path(model_dir)
    
    # Load configuration
    with open(model_dir / 'action_model_config.pkl', 'rb') as f:
        config = pickle.load(f)
    
    # Create model
    model = ActionRecognitionModel(
        num_classes=config['num_classes'],
        dropout=config['dropout']
    )
    
    # Load weights
    model.load_state_dict(torch.load(model_dir / 'action_model_final.pth',
                                     map_location=device))
    model = model.to(device)
    model.eval()
    
    return model