File size: 11,489 Bytes
0966609 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 |
import os
import torch
import torch.nn as nn
import torch.optim as optim
from torch.utils.data import DataLoader
from tqdm import tqdm
import random
import ssl
# Disable SSL verification for downloading pretrained weights
ssl._create_default_https_context = ssl._create_unverified_context
from torch.cuda.amp import GradScaler, autocast
from src.config import Config
from src.models import DeepfakeDetector
from src.dataset import DeepfakeDataset
try:
from safetensors.torch import save_file, load_file
SAFETENSORS_AVAILABLE = True
except ImportError:
SAFETENSORS_AVAILABLE = False
print("Warning: safetensors not installed. Checkpoints will be saved as .pt")
def train():
# Setup
Config.setup()
device = torch.device(Config.DEVICE)
# --- Data Loading with Automatic Split ---
if Config.TRAIN_DATA_PATH == Config.TEST_DATA_PATH:
print("Train and Test paths are identical. Performing automatic 80/20 shuffle split...")
all_paths, all_labels = DeepfakeDataset.scan_directory(Config.TRAIN_DATA_PATH)
if len(all_paths) == 0:
print(f"No images found in {Config.TRAIN_DATA_PATH}")
return
# Combine and shuffle
combined = list(zip(all_paths, all_labels))
random.shuffle(combined)
split_idx = int(len(combined) * 0.8)
train_data = combined[:split_idx]
val_data = combined[split_idx:]
train_paths, train_labels = zip(*train_data)
val_paths, val_labels = zip(*val_data)
train_dataset = DeepfakeDataset(file_paths=list(train_paths), labels=list(train_labels), phase='train')
val_dataset = DeepfakeDataset(file_paths=list(val_paths), labels=list(val_labels), phase='val')
else:
# Standard folder-based loading
train_dataset = DeepfakeDataset(root_dir=Config.TRAIN_DATA_PATH, phase='train')
val_dataset = DeepfakeDataset(root_dir=Config.TEST_DATA_PATH, phase='val')
# Dataloaders
train_loader = DataLoader(train_dataset, batch_size=Config.BATCH_SIZE, shuffle=True,
num_workers=Config.NUM_WORKERS,
pin_memory=True if device.type=='cuda' else False,
persistent_workers=True if Config.NUM_WORKERS > 0 else False)
val_loader = DataLoader(val_dataset, batch_size=Config.BATCH_SIZE, shuffle=False,
num_workers=Config.NUM_WORKERS,
pin_memory=True if device.type=='cuda' else False,
persistent_workers=True if Config.NUM_WORKERS > 0 else False)
# Model
print("Initializing Multi-Branch DeepfakeDetector...")
model = DeepfakeDetector(pretrained=True).to(device)
# Optimization
criterion = nn.BCEWithLogitsLoss()
optimizer = optim.AdamW(model.parameters(), lr=Config.LEARNING_RATE, weight_decay=Config.WEIGHT_DECAY)
# Optimization
criterion = nn.BCEWithLogitsLoss()
optimizer = optim.AdamW(model.parameters(), lr=Config.LEARNING_RATE, weight_decay=Config.WEIGHT_DECAY)
scheduler = optim.lr_scheduler.StepLR(optimizer, step_size=5, gamma=0.5)
# Enable AMP only for CUDA (Windows NVIDIA)
use_amp = (Config.DEVICE == 'cuda')
scaler = GradScaler() if use_amp else None
if use_amp:
print("π Mixed Precision (AMP) Enabled for RTX GPU")
else:
print("π Standard Precision (No AMP) for CPU/MPS")
# Resume from checkpoint if exists
start_epoch = 0
best_acc = 0.0
# Priority:
# 1. best_model.safetensors (if we crashed mid-training)
# 2. patched_model.safetensors (the model we want to improve)
resume_path = os.path.join(Config.CHECKPOINT_DIR, "best_model.safetensors")
if not os.path.exists(resume_path):
# Look for latest epoch checkpoint
import glob
import re
checkpoints = glob.glob(os.path.join(Config.CHECKPOINT_DIR, "checkpoint_ep*.safetensors"))
if checkpoints:
# Sort by epoch number
def get_epoch(p):
match = re.search(r"checkpoint_ep(\d+)", p)
return int(match.group(1)) if match else 0
latest_ckpt = max(checkpoints, key=get_epoch)
resume_path = latest_ckpt
start_epoch = get_epoch(latest_ckpt)
print(f"π Auto-Resuming from latest epoch: {start_epoch}")
else:
resume_path = os.path.join(Config.CHECKPOINT_DIR, "patched_model.safetensors")
if os.path.exists(resume_path):
print(f"\nπ Found existing checkpoint: {resume_path}")
print("Auto-resuming to FINETUNE this model...")
try:
if resume_path.endswith(".safetensors") and SAFETENSORS_AVAILABLE:
state_dict = load_file(resume_path)
else:
state_dict = torch.load(resume_path, map_location=device)
# Use strict=False to allow for minor architecture changes or missing keys
model.load_state_dict(state_dict, strict=False)
print("β
Weights loaded. Starting Fine-Tuning.")
except Exception as e:
print(f"β Failed to load checkpoint: {e}")
print("Starting from ImageNet weights.")
# Loop
for epoch in range(start_epoch, Config.EPOCHS):
model.train()
train_loss = 0.0
train_correct = 0
train_total = 0
loop = tqdm(train_loader, desc=f"Epoch {epoch+1}/{Config.EPOCHS}")
for images, labels in loop:
images = images.to(device)
labels = labels.to(device).unsqueeze(1)
optimizer.zero_grad()
if use_amp:
with autocast():
outputs = model(images)
loss = criterion(outputs, labels)
scaler.scale(loss).backward()
scaler.step(optimizer)
scaler.update()
else:
# Standard training for Mac/CPU
outputs = model(images)
loss = criterion(outputs, labels)
loss.backward()
optimizer.step()
train_loss += loss.item()
preds = (torch.sigmoid(outputs) > 0.5).float()
correct = (preds == labels).sum().item()
train_correct += correct
train_total += labels.size(0)
loop.set_postfix(loss=loss.item(), acc=correct/labels.size(0))
train_acc = train_correct / train_total if train_total > 0 else 0
print(f"Epoch {epoch+1} Train Loss: {train_loss/len(train_loader):.4f} Acc: {train_acc:.4f}")
# Save checkpoint after every epoch
save_checkpoint(model, epoch+1, train_acc, best=False)
# Validation
if len(val_dataset) > 0:
val_loss, val_acc = validate(model, val_loader, criterion, device)
print(f"Epoch {epoch+1} Val Loss: {val_loss:.4f} Acc: {val_acc:.4f}")
# Save best model if validation accuracy improved
if val_acc > best_acc:
best_acc = val_acc
print(f"β New best model! Validation Accuracy: {val_acc:.4f}")
save_checkpoint(model, epoch+1, val_acc, best=True)
scheduler.step()
print(f"\nπ Training Complete!")
print(f"Best Validation Accuracy: {best_acc:.4f}")
def validate(model, loader, criterion, device):
model.eval()
val_loss = 0.0
correct = 0
total = 0
with torch.no_grad():
for images, labels in loader:
images = images.to(device)
labels = labels.to(device).unsqueeze(1)
outputs = model(images)
loss = criterion(outputs, labels)
val_loss += loss.item()
preds = (torch.sigmoid(outputs) > 0.5).float()
correct += (preds == labels).sum().item()
total += labels.size(0)
return val_loss / len(loader), correct / total
def save_checkpoint(model, epoch, acc, best=False):
state_dict = model.state_dict()
name = "best_model.safetensors" if best else f"checkpoint_ep{epoch}.safetensors"
path = os.path.join(Config.CHECKPOINT_DIR, name)
if SAFETENSORS_AVAILABLE:
try:
# Try with shared tensors support
from safetensors.torch import save_model
save_model(model, path)
print(f"Saved Checkpoint: {path}")
# π Auto-Log to History
try:
from datetime import datetime
log_path = os.path.join(Config.PROJECT_ROOT, "TRAINING_HISTORY.md")
timestamp = datetime.now().strftime("%Y-%m-%d | %I:%M %p")
# Create file with header if doesn't exist
if not os.path.exists(log_path):
with open(log_path, "w", encoding="utf-8") as f:
f.write("# π Training History Log\n\n")
f.write("| Date | Time | Model Name | Dataset | Epochs | Accuracy | Loss | Status |\n")
f.write("| :--- | :--- | :--- | :--- | :--- | :--- | :--- | :--- |\n")
# Append Entry to Summary Log
with open(log_path, "a", encoding="utf-8") as f:
# Format: Date | Time | Name | Dataset | Epoch | Acc | Loss | Status
dataset_name = os.path.basename(Config.DATA_DIR)
entry = f"| **{timestamp.split(' | ')[0]}** | {timestamp.split(' | ')[1]} | {name} | {dataset_name} | {epoch} | {acc*100:.2f}% | N/A | β
Saved |\n"
f.write(entry)
print(f"π Logged to TRAINING_HISTORY.md")
# π Detailed Lab Notebook Logging
detail_path = os.path.join(Config.PROJECT_ROOT, "DETAILED_HISTORY.md")
with open(detail_path, "a", encoding="utf-8") as f:
f.write(f"\n## Model: {name} (Epoch {epoch})\n")
f.write(f"| Feature | Detail |\n| :--- | :--- |\n")
f.write(f"| **Date** | {timestamp} |\n")
f.write(f"| **Training Accuracy** | {acc*100:.2f}% |\n")
f.write(f"| **Dataset** | {Config.DATA_DIR} |\n")
f.write(f"| **Batch Size** | {Config.BATCH_SIZE} |\n")
f.write(f"| **Optimizer** | AdamW (lr={Config.LEARNING_RATE}) |\n")
f.write(f"| **Device** | {Config.DEVICE.upper()} |\n")
f.write("\n---\n")
print(f"π Detailed log written to DETAILED_HISTORY.md")
except Exception as e:
print(f"β οΈ Failed to write log: {e}")
except Exception as e:
# Fallback to regular torch save if safetensors fails
print(f"SafeTensors save failed ({e}), falling back to .pth format")
torch.save(state_dict, path.replace(".safetensors", ".pth"))
print(f"Saved Checkpoint (Legacy): {path.replace('.safetensors', '.pth')}")
else:
torch.save(state_dict, path.replace(".safetensors", ".pth"))
print(f"Saved Checkpoint (Legacy): {path}")
if __name__ == "__main__":
train()
|