Muzaffar branch

deepfake-inceptionv4.ipynb

@@ -0,0 +1 @@
+{"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"name":"python","version":"3.11.13","mimetype":"text/x-python","codemirror_mode":{"name":"ipython","version":3},"pygments_lexer":"ipython3","nbconvert_exporter":"python","file_extension":".py"},"kaggle":{"accelerator":"gpu","dataSources":[{"sourceId":8497982,"sourceType":"datasetVersion","datasetId":5070657}],"dockerImageVersionId":31090,"isInternetEnabled":true,"language":"python","sourceType":"notebook","isGpuEnabled":true}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"code","source":"import os\nimport torch\nimport timm\nimport torch.nn as nn\nimport torch.optim as optim\nfrom torch.utils.data import DataLoader\nfrom torchvision import datasets, transforms\nfrom PIL import ImageFile, Image\nimport matplotlib.pyplot as plt\n\nImageFile.LOAD_TRUNCATED_IMAGES = True\nos.environ[\"KMP_DUPLICATE_LIB_OK\"] = \"TRUE\"\n\nclass SafeImageFolder(datasets.ImageFolder):\n    def __getitem__(self, index):\n        try:\n            path, target = self.imgs[index]\n            img = Image.open(path)\n            if img.mode != 'RGB':\n                img = img.convert('RGB')\n            if self.transform is not None:\n                img = self.transform(img)\n            return img, target\n        except (OSError, IOError):\n            print(f\"[Warning] Skipping corrupted image at index {index}\")\n            return self.__getitem__((index + 1) % len(self.imgs))\n\ntrain_dir = \"/kaggle/input/ai-generated-images-vs-real-images/train\"\ntest_dir = \"/kaggle/input/ai-generated-images-vs-real-images/test\"\n\nimg_size = 299  # InceptionV4 default input\nbatch_size = 32\n\ntransform = transforms.Compose([\n    transforms.Resize((img_size, img_size)),\n    transforms.ToTensor(),\n    transforms.Normalize([0.5], [0.5])  # scale to [-1, 1]\n])\n\n\nif __name__ == '__main__':\n    train_data = SafeImageFolder(train_dir, transform=transform)\n    test_data = SafeImageFolder(test_dir, transform=transform)\n    print(f\"Found {len(train_data)} training images and {len(test_data)} test images.\")\n    if len(train_data) == 0:\n        raise ValueError(f\"No images found in {train_dir}. Please check your data path and structure.\")\n    if len(test_data) == 0:\n        raise ValueError(f\"No images found in {test_dir}. Please check your data path and structure.\")\n\n    train_loader = DataLoader(train_data, batch_size=batch_size, shuffle=True, num_workers=0)\n    test_loader = DataLoader(test_data, batch_size=batch_size, shuffle=False, num_workers=0)\n\n    device = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\n\n    model = timm.create_model(\"inception_v4\", pretrained=True, num_classes=2)\n    model = model.to(device)\n\n    criterion = nn.CrossEntropyLoss()\n    optimizer = optim.Adam(model.parameters(), lr=1e-4)\n\n    epochs = 5\n    train_loss_hist, test_loss_hist = [], []\n    train_acc_hist, test_acc_hist = [], []\n\n    for epoch in range(epochs):\n        model.train()\n        running_loss = 0.0\n        running_corrects = torch.tensor(0, dtype=torch.float32, device=device)\n        for images, labels in train_loader:\n            images, labels = images.to(device), labels.to(device)\n\n            optimizer.zero_grad()\n            outputs = model(images)\n            loss = criterion(outputs, labels)\n\n            _, preds = torch.max(outputs, 1)\n            loss.backward()\n            optimizer.step()\n\n            running_loss += loss.item() * images.size(0)\n            running_corrects += torch.sum(preds == labels.data)\n\n        epoch_loss = running_loss / len(train_data)\n        epoch_acc = running_corrects.double() / len(train_data)\n        train_loss_hist.append(epoch_loss)\n        train_acc_hist.append(epoch_acc.item())\n\n        model.eval()\n        val_loss = 0.0\n        val_corrects = torch.tensor(0, dtype=torch.float32, device=device)\n        with torch.no_grad():\n            for images, labels in test_loader:\n                images, labels = images.to(device), labels.to(device)\n                outputs = model(images)\n                loss = criterion(outputs, labels)\n\n                _, preds = torch.max(outputs, 1)\n                val_loss += loss.item() * images.size(0)\n                val_corrects += torch.sum(preds == labels.data)\n\n        val_loss /= len(test_data)\n        val_acc = val_corrects.double() / len(test_data)\n        test_loss_hist.append(val_loss)\n        test_acc_hist.append(val_acc.item())\n\n        print(f\"Epoch {epoch+1}/{epochs} - \"\n              f\"Train Loss: {epoch_loss:.4f}, Train Acc: {epoch_acc:.4f} - \"\n              f\"Val Loss: {val_loss:.4f}, Val Acc: {val_acc:.4f}\")\n\n    torch.save(model.state_dict(), \"inceptionv4_deepfake.pth\")\n    print(\"Model weights saved as inceptionv4_deepfake.pth\")\n\n    plt.figure(figsize=(10,5))\n    plt.plot(train_loss_hist, label=\"Train Loss\")\n    plt.plot(test_loss_hist, label=\"Val Loss\")\n    plt.xlabel(\"Epoch\")\n    plt.ylabel(\"Loss\")\n    plt.legend()\n    plt.title(\"Training & Validation Loss\")\n    plt.show()\n\n    plt.figure(figsize=(10,5))\n    plt.plot(train_acc_hist, label=\"Train Acc\")\n    plt.plot(test_acc_hist, label=\"Val Acc\")\n    plt.xlabel(\"Epoch\")\n    plt.ylabel(\"Accuracy\")\n    plt.legend()\n    plt.title(\"Training & Validation Accuracy\")\n    plt.savefig(\"training_curves.png\")\n    plt.show()\n    print(\"training_curves.png saved\")\n","metadata":{"_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","trusted":true,"execution":{"iopub.status.busy":"2025-08-28T03:44:15.239735Z","iopub.execute_input":"2025-08-28T03:44:15.240097Z","execution_failed":"2025-08-28T14:52:49.156Z"}},"outputs":[{"name":"stdout","text":"Found 48000 training images and 12000 test images.\n","output_type":"stream"},{"name":"stderr","text":"/usr/local/lib/python3.11/dist-packages/PIL/Image.py:3442: DecompressionBombWarning: Image size (98058240 pixels) exceeds limit of 89478485 pixels, could be decompression bomb DOS attack.\n  warnings.warn(\n/usr/local/lib/python3.11/dist-packages/PIL/Image.py:3442: DecompressionBombWarning: Image size (99991727 pixels) exceeds limit of 89478485 pixels, could be decompression bomb DOS attack.\n  warnings.warn(\n/usr/local/lib/python3.11/dist-packages/PIL/Image.py:1043: UserWarning: Palette images with Transparency expressed in bytes should be converted to RGBA images\n  warnings.warn(\n/usr/local/lib/python3.11/dist-packages/PIL/Image.py:3442: DecompressionBombWarning: Image size (98806617 pixels) exceeds limit of 89478485 pixels, could be decompression bomb DOS attack.\n  warnings.warn(\n/usr/local/lib/python3.11/dist-packages/PIL/Image.py:3442: DecompressionBombWarning: Image size (96000000 pixels) exceeds limit of 89478485 pixels, could be decompression bomb DOS attack.\n  warnings.warn(\n/usr/local/lib/python3.11/dist-packages/PIL/Image.py:3442: DecompressionBombWarning: Image size (107184040 pixels) exceeds limit of 89478485 pixels, could be decompression bomb DOS attack.\n  warnings.warn(\n/usr/local/lib/python3.11/dist-packages/PIL/Image.py:3442: DecompressionBombWarning: Image size (96012000 pixels) exceeds limit of 89478485 pixels, could be decompression bomb DOS attack.\n  warnings.warn(\n/usr/local/lib/python3.11/dist-packages/PIL/Image.py:3442: DecompressionBombWarning: Image size (90671520 pixels) exceeds limit of 89478485 pixels, could be decompression bomb DOS attack.\n  warnings.warn(\n/usr/local/lib/python3.11/dist-packages/PIL/Image.py:3442: DecompressionBombWarning: Image size (161087488 pixels) exceeds limit of 89478485 pixels, could be decompression bomb DOS attack.\n  warnings.warn(\n/usr/local/lib/python3.11/dist-packages/PIL/Image.py:3442: DecompressionBombWarning: Image size (143040000 pixels) exceeds limit of 89478485 pixels, could be decompression bomb DOS attack.\n  warnings.warn(\n/usr/local/lib/python3.11/dist-packages/PIL/Image.py:3442: DecompressionBombWarning: Image size (121554000 pixels) exceeds limit of 89478485 pixels, could be decompression bomb DOS attack.\n  warnings.warn(\n","output_type":"stream"},{"name":"stdout","text":"Epoch 1/10 - Train Loss: 0.1990, Train Acc: 0.9209 - Val Loss: 0.1253, Val Acc: 0.9542\nEpoch 2/10 - Train Loss: 0.1057, Train Acc: 0.9610 - Val Loss: 0.1006, Val Acc: 0.9629\nEpoch 3/10 - Train Loss: 0.0692, Train Acc: 0.9740 - Val Loss: 0.1122, Val Acc: 0.9586\nEpoch 4/10 - Train Loss: 0.0491, Train Acc: 0.9821 - Val Loss: 0.1206, Val Acc: 0.9563\nEpoch 5/10 - Train Loss: 0.0440, Train Acc: 0.9839 - Val Loss: 0.1471, Val Acc: 0.9544\n","output_type":"stream"}],"execution_count":null},{"cell_type":"code","source":"","metadata":{"trusted":true},"outputs":[],"execution_count":null}]}