Dog breed identification. This means training a model on pictures of dogs. Then letting it recognize test pictures. The rate of correctly identified pictures measures the success.
\n\n**Related Works**\nCNN + SVM (Stanford Dogs, 2024)
\nCui, Y. et al. (2024). Classification of Dog Breeds Using CNN Models and SVM. Bioengineering, 11(11), 1157. doi:10.3390/bioengineering11111157\nMDPI
Hybrid InceptionV3-Xception (2025)
\nSai Ram, G. (2025). Next-Gen dog breed identification using hybrid and custom CNN models. Int J Comput Artif Intell, 6(1):99–110. doi:10.33545/27076571.2025.v6.i1b.138\nComputersciencejournals
= Taking a model that has been trained on a similar dataset and adjusting it to fit our problem
\nhttps://keras.io/guides/transfer_learning/#transfer-learning-amp-finetuning
\n\n1. Instantiate a base model and load pre-trained weights into it.\n2. Freeze all layers in the base model by setting trainable = False.\n3. Create a new model on top of the output of one (or several) layers from the base model.\n4. Train your new model on your new dataset.\n\nOptional: Fine-tuning the model. This is done by unfreezing layers of the base model and retraining it with a low learning rate
","metadata":{}},{"cell_type":"code","source":"from tensorflow.keras import Sequential, Model\nfrom tensorflow.keras.layers import Input, Conv2D, MaxPooling2D, GlobalAveragePooling2D, Flatten, Dense\n\ndef get_model(model_name):\n if model_name == \"seq\":\n # Create sequential model\n # Each part is a layer: Input = Flatten -> Hidden = Dense -> Output = Dense\n # No convolutional part\n return Sequential([\n Flatten(input_shape=(299, 299, 3)), # Flattened image input\n Dense(128, activation=\"relu\"), # Hidden layer, 128 neurons with ReLU function\n Dense(120) # 120 output classes\n ])\n elif model_name == \"xception\":\n # Create Inception base model\n base_model = tf.keras.applications.Xception(\n #input_shape = (299, 299 , 3), # Defaults to (299, 299, 3) if include_top is true\n include_top = False, # If false, last layer (output) is excluded\n weights = \"imagenet\", # Pre-trained on imagenet\n #classes = 120, # Output classes, works only if include_top = True and weights = None\n #classifier_activation = \"softmax\", # Activation function in output layer\n )\n \n # Freeze the base model layers -> They won't get trained\n base_model.trainable = False\n \n # Define shape of the input\n input_tensor = Input(shape = IMAGE_DIMENSIONS)\n \n # Build layers on top of base_model\n x = base_model(input_tensor, training = False)\n x = GlobalAveragePooling2D()(x)\n x = Dense(1024, activation=\"relu\")(x)\n outputs = Dense(OUTPUT_SIZE, activation=\"softmax\")(x) # Add output layer with 120 labels\n\n # Create new model by combining input and output\n model = Model(inputs = input_tensor, outputs = outputs)\n\n return model\n\nmodel = get_model(\"xception\")","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2026-06-09T20:09:25.764246Z","iopub.execute_input":"2026-06-09T20:09:25.764557Z","iopub.status.idle":"2026-06-09T20:09:32.294880Z","shell.execute_reply.started":"2026-06-09T20:09:25.764534Z","shell.execute_reply":"2026-06-09T20:09:32.294034Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"## Adjust model settings\nmodel.compile(\n optimizer=\"adam\", # Using Adam algorithm\n loss=tf.keras.losses.CategoricalCrossentropy(from_logits=True), # Loss function\n metrics=[\"accuracy\"] # Metrics - Amount of correctly classified images\n)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2026-06-09T20:09:32.295829Z","iopub.execute_input":"2026-06-09T20:09:32.296148Z","iopub.status.idle":"2026-06-09T20:09:32.308496Z","shell.execute_reply.started":"2026-06-09T20:09:32.296126Z","shell.execute_reply":"2026-06-09T20:09:32.307818Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"model.summary()","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"## Train the model for 10 iterations (fit model to the data)\nmodel.fit(\n X_train, \n y_train, \n epochs=10\n)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2026-06-09T20:09:32.309379Z","iopub.execute_input":"2026-06-09T20:09:32.309630Z","iopub.status.idle":"2026-06-09T20:15:35.505812Z","shell.execute_reply.started":"2026-06-09T20:09:32.309611Z","shell.execute_reply":"2026-06-09T20:15:35.505188Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"## Evaluate the accuracy\ntest_loss, test_acc = model.evaluate(X_val, y_val, verbose=2)\nprint(f\"Test accuracy: {test_acc}\")","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2026-06-09T20:15:35.506800Z","iopub.execute_input":"2026-06-09T20:15:35.507313Z","iopub.status.idle":"2026-06-09T20:16:18.108760Z","shell.execute_reply.started":"2026-06-09T20:15:35.507289Z","shell.execute_reply":"2026-06-09T20:16:18.108098Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"## Save model to working directory\nimport datetime\n\ndef save_model(model):\n working_dir = \"/kaggle/working/\"\n timestamp = datetime.datetime.now().strftime(\"%Y%m%d-%H%M%S\")\n filepath = f\"{working_dir}model-{timestamp}.keras\"\n \n model.save(filepath)\n \n print(f\"Saved model at {filepath}\")\n\nsave_model(model)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2026-06-09T20:25:22.289262Z","iopub.execute_input":"2026-06-09T20:25:22.289888Z","iopub.status.idle":"2026-06-09T20:25:22.987194Z","shell.execute_reply.started":"2026-06-09T20:25:22.289860Z","shell.execute_reply":"2026-06-09T20:25:22.986397Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"## Load a saved model\ndef load_model(path):\n model = tf.keras.models.load_model(path)\n print(f\"Loaded model from {path}\")\n return model","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"