File size: 4,746 Bytes
40243b5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import torch
import torch.nn as nn
import torchvision.models as models
import sys
import os
import pickle
import re
from collections import Counter

DEVICE = torch.device("cuda" if torch.cuda.is_available() else "cpu")

EMBED_DIM = 512
HIDDEN_DIM = 512
MAX_LEN = 25

# Vocabulary class
class Vocabulary:
    def __init__(self, freq_threshold=5):
        self.freq_threshold = freq_threshold
        self.itos = {0: "pad", 1: "startofseq", 2: "endofseq", 3: "unk"}
        self.stoi = {v: k for k, v in self.itos.items()}
        self.index = 4

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

    def tokenizer(self, text):
        text = text.lower()
        tokens = re.findall(r"\w+", text)
        return tokens

    def build_vocabulary(self, sentence_list):
        frequencies = Counter()
        for sentence in sentence_list:
            tokens = self.tokenizer(sentence)
            frequencies.update(tokens)

        for word, freq in frequencies.items():
            if freq >= self.freq_threshold:
                self.stoi[word] = self.index
                self.itos[self.index] = word
                self.index += 1

    def numericalize(self, text):
        tokens = self.tokenizer(text)
        numericalized = []
        for token in tokens:
            if token in self.stoi:
                numericalized.append(self.stoi[token])
            else:
                numericalized.append(self.stoi["unk"])
        return numericalized


class Encoder(nn.Module):
    def __init__(self, embed_dim):
        super().__init__()
        resnet = models.resnet50(weights=models.ResNet50_Weights.DEFAULT)
        self.backbone = nn.Sequential(*list(resnet.children())[:-1])
        self.fc = nn.Linear(resnet.fc.in_features, embed_dim)
        self.bn = nn.BatchNorm1d(embed_dim)

    def forward(self, x):
        with torch.no_grad():
            features = self.backbone(x)
        features = features.reshape(features.size(0), -1)
        features = self.bn(self.fc(features))
        return features


class Decoder(nn.Module):
    def __init__(self, embed_dim, hidden_dim, vocab_size):
        super().__init__()
        self.embedding = nn.Embedding(vocab_size, embed_dim)
        self.lstm = nn.LSTM(
            embed_dim,
            hidden_dim,
            batch_first=True
        )
        self.fc = nn.Linear(hidden_dim, vocab_size)

    def forward(self, x, states=None):
        emb = self.embedding(x)
        outputs, states = self.lstm(emb, states)
        logits = self.fc(outputs)
        return logits, states


class CaptionModel(nn.Module):
    def __init__(self, embed_dim, hidden_dim, vocab_size):
        super().__init__()
        self.encoder = Encoder(embed_dim)
        self.decoder = Decoder(embed_dim, hidden_dim, vocab_size)


# Main debug
script_dir = os.path.dirname(os.path.abspath(__file__))
CHECKPOINT_PATH = os.path.join(script_dir, "best_checkpoint.pth")
VOCAB_PATH = os.path.join(script_dir, "vocab.pkl")

print("=" * 80)
print("LOADING CHECKPOINT")
print("=" * 80)

checkpoint = torch.load(CHECKPOINT_PATH, map_location=DEVICE)
print(f"\nCheckpoint keys: {list(checkpoint.keys())}")

print("\nCheckpoint model_state_dict keys:")
checkpoint_keys = set(checkpoint["model_state_dict"].keys())
for key in sorted(checkpoint_keys):
    shape = checkpoint["model_state_dict"][key].shape
    print(f"  {key}: {shape}")

# Load vocab
with open(VOCAB_PATH, "rb") as f:
    vocab = pickle.load(f)

vocab_size = len(vocab)
print(f"\nVocab size: {vocab_size}")

# Create model
model = CaptionModel(
    EMBED_DIM,
    HIDDEN_DIM,
    vocab_size
).to(DEVICE)

print("\n" + "=" * 80)
print("MODEL STATE DICT KEYS")
print("=" * 80)

model_keys = set(model.state_dict().keys())
for key in sorted(model_keys):
    shape = model.state_dict()[key].shape
    print(f"  {key}: {shape}")

# Check differences
print("\n" + "=" * 80)
print("COMPARISON")
print("=" * 80)

print("\nKeys in checkpoint but NOT in model:")
for key in sorted(checkpoint_keys - model_keys):
    print(f"  {key}")

print("\nKeys in model but NOT in checkpoint:")
for key in sorted(model_keys - checkpoint_keys):
    print(f"  {key}")

print("\nKeys in both but with different shapes:")
for key in sorted(checkpoint_keys & model_keys):
    cp_shape = checkpoint["model_state_dict"][key].shape
    model_shape = model.state_dict()[key].shape
    if cp_shape != model_shape:
        print(f"  {key}")
        print(f"    Checkpoint:  {cp_shape}")
        print(f"    Model:       {model_shape}")

print("\n" + "=" * 80)
print("ATTEMPTING TO LOAD WEIGHTS")
print("=" * 80)

try:
    model.load_state_dict(checkpoint["model_state_dict"])
    print("SUCCESS: Weights loaded successfully!")
except Exception as e:
    print(f"ERROR: {e}")