diff --git a/.github/workflows/test_and_deploy.yml b/.github/workflows/test_and_deploy.yml
index 78be7bb5..0b51bb58 100644
--- a/.github/workflows/test_and_deploy.yml
+++ b/.github/workflows/test_and_deploy.yml
@@ -37,12 +37,15 @@ jobs:
     name: Run package tests
     timeout-minutes: 60
     runs-on: ${{ matrix.os }}
+    env:
+      KERAS_BACKEND: torch
+      CELLFINDER_TEST_DEVICE: cpu
     strategy:
       matrix:
         # Run all supported Python versions on linux
         os: [ubuntu-latest]
-        python-version: ["3.9", "3.10"]
-        # Include one windows, one macos run each for M1 (latest) and Intel (13)
+        python-version: ["3.9", "3.10", "3.11"]
+        # Include one windows and two macOS (intel based and arm based) runs
         include:
           - os: macos-13
             python-version: "3.10"
@@ -80,11 +83,13 @@ jobs:
        NUMBA_DISABLE_JIT: "1"
 
     steps:
-      - name: Cache tensorflow model
+      - name: Cache brainglobe directory
         uses: actions/cache@v3
         with:
-          path: "~/.cellfinder"
-          key: models-${{ hashFiles('~/.brainglobe/**') }}
+          path: | # ensure we don't cache any interrupted atlas download and extraction, if e.g. we cancel the workflow manually
+            ~/.brainglobe
+            !~/.brainglobe/atlas.tar.gz
+          key: brainglobe
       # Setup pyqt libraries
       - name: Setup qtpy libraries
         uses: tlambert03/setup-qt-libs@v1
@@ -104,13 +109,17 @@ jobs:
     name: Run brainmapper tests to check for breakages
     timeout-minutes: 60
     runs-on: ubuntu-latest
+    env:
+      KERAS_BACKEND: torch
+      CELLFINDER_TEST_DEVICE: cpu
     steps:
-      - name: Cache tensorflow model
+      - name: Cache brainglobe directory
         uses: actions/cache@v3
         with:
-          path: "~/.cellfinder"
-          key: models-${{ hashFiles('~/.brainglobe/**') }}
-
+          path: | # ensure we don't cache any interrupted atlas download and extraction, if e.g. we cancel the workflow manually
+            ~/.brainglobe
+            !~/.brainglobe/atlas.tar.gz
+          key: brainglobe
       - name: Checkout brainglobe-workflows
         uses: actions/checkout@v3
         with:
@@ -124,8 +133,9 @@ jobs:
       - name: Install test dependencies
         run: |
           python -m pip install --upgrade pip wheel
-          # Install latest SHA on this brainglobe-workflows branch
-          python -m pip install git+$GITHUB_SERVER_URL/$GITHUB_REPOSITORY@$GITHUB_SHA
+          # Install cellfinder from the latest SHA on this branch
+          python -m pip install "cellfinder @ git+$GITHUB_SERVER_URL/$GITHUB_REPOSITORY@$GITHUB_SHA"
+
           # Install checked out copy of brainglobe-workflows
           python -m pip install .[dev]
 
diff --git a/.github/workflows/test_include_guard.yaml b/.github/workflows/test_include_guard.yaml
index 70415a2e..07a5eda8 100644
--- a/.github/workflows/test_include_guard.yaml
+++ b/.github/workflows/test_include_guard.yaml
@@ -1,5 +1,5 @@
-name: Test Tensorflow include guards
-# These tests check that the include guards checking for tensorflow's availability
+name: Test Keras include guards
+# These tests check that the include guards checking for Keras availability
 # behave as expected on ubuntu and macOS.
 
 on:
@@ -9,7 +9,7 @@ on:
       - main
 
 jobs:
-  tensorflow_guards:
+  keras_guards:
     name: Test include guards
     strategy:
       matrix:
@@ -24,24 +24,21 @@ jobs:
         with:
           python-version: '3.10'
 
-      - name: Install via pip
-        run: python -m pip install -e .
+      - name: Install cellfinder via pip
+        run: python -m pip install -e "."
 
       - name: Test (working) import
         uses: jannekem/run-python-script-action@v1
+        env:
+          KERAS_BACKEND: torch
         with:
           fail-on-error: true
           script: |
             import cellfinder.core
             import cellfinder.napari
 
-      - name: Uninstall tensorflow-macos on Mac M1
-        if: matrix.os == 'macos-latest'
-        run: python -m pip uninstall -y tensorflow-macos
-
-      - name: Uninstall tensorflow on Ubuntu
-        if: matrix.os == 'ubuntu-latest'
-        run: python -m pip uninstall -y tensorflow
+      - name: Uninstall keras
+        run: python -m pip uninstall -y keras
 
       - name: Test (broken) import
         id: broken_import
diff --git a/cellfinder/__init__.py b/cellfinder/__init__.py
index fcd51af8..405524c9 100644
--- a/cellfinder/__init__.py
+++ b/cellfinder/__init__.py
@@ -1,25 +1,31 @@
+import os
 from importlib.metadata import PackageNotFoundError, version
 from pathlib import Path
 
+# Check cellfinder is installed
 try:
     __version__ = version("cellfinder")
 except PackageNotFoundError as e:
     raise PackageNotFoundError("cellfinder package not installed") from e
 
-# If tensorflow is not present, tools cannot be used.
+# If Keras is not present, tools cannot be used.
 # Throw an error in this case to prevent invocation of functions.
 try:
-    TF_VERSION = version("tensorflow")
+    KERAS_VERSION = version("keras")
 except PackageNotFoundError as e:
-    try:
-        TF_VERSION = version("tensorflow-macos")
-    except PackageNotFoundError as e:
-        raise PackageNotFoundError(
-            f"cellfinder tools cannot be invoked without tensorflow. "
-            f"Please install tensorflow into your environment to use cellfinder tools. "
-            f"For more information, please see "
-            f"https://github.com/brainglobe/brainglobe-meta#readme."
-        ) from e
+    raise PackageNotFoundError(
+        f"cellfinder tools cannot be invoked without Keras. "
+        f"Please install Keras with a backend into your environment "
+        f"to use cellfinder tools. "
+        f"For more information on Keras backends, please see "
+        f"https://keras.io/getting_started/#installing-keras-3."
+        f"For more information on brainglobe, please see "
+        f"https://github.com/brainglobe/brainglobe-meta#readme."
+    ) from e
+
+
+# Set the Keras backend to torch
+os.environ["KERAS_BACKEND"] = "torch"
 
 __author__ = "Adam Tyson, Christian Niedworok, Charly Rousseau"
 __license__ = "BSD-3-Clause"
diff --git a/cellfinder/core/classify/classify.py b/cellfinder/core/classify/classify.py
index 86b709f5..1da4dabe 100644
--- a/cellfinder/core/classify/classify.py
+++ b/cellfinder/core/classify/classify.py
@@ -1,10 +1,10 @@
 import os
 from typing import Any, Callable, Dict, List, Optional, Tuple
 
+import keras
 import numpy as np
 from brainglobe_utils.cells.cells import Cell
 from brainglobe_utils.general.system import get_num_processes
-from tensorflow import keras
 
 from cellfinder.core import logger, types
 from cellfinder.core.classify.cube_generator import CubeGeneratorFromFile
@@ -48,9 +48,7 @@ def main(
         callbacks = None
 
     # Too many workers doesn't increase speed, and uses huge amounts of RAM
-    workers = get_num_processes(
-        min_free_cpu_cores=n_free_cpus, n_max_processes=max_workers
-    )
+    workers = get_num_processes(min_free_cpu_cores=n_free_cpus)
 
     logger.debug("Initialising cube generator")
     inference_generator = CubeGeneratorFromFile(
@@ -63,6 +61,8 @@ def main(
         cube_width=cube_width,
         cube_height=cube_height,
         cube_depth=cube_depth,
+        use_multiprocessing=False,
+        workers=workers,
     )
 
     model = get_model(
@@ -73,10 +73,9 @@ def main(
     )
 
     logger.info("Running inference")
+    # in Keras 3.0 multiprocessing params are specified in the generator
     predictions = model.predict(
         inference_generator,
-        use_multiprocessing=True,
-        workers=workers,
         verbose=True,
         callbacks=callbacks,
     )
diff --git a/cellfinder/core/classify/cube_generator.py b/cellfinder/core/classify/cube_generator.py
index 56f226cd..4a24467f 100644
--- a/cellfinder/core/classify/cube_generator.py
+++ b/cellfinder/core/classify/cube_generator.py
@@ -2,13 +2,13 @@
 from random import shuffle
 from typing import Dict, List, Optional, Tuple, Union
 
+import keras
 import numpy as np
-import tensorflow as tf
 from brainglobe_utils.cells.cells import Cell, group_cells_by_z
 from brainglobe_utils.general.numerical import is_even
+from keras.utils import Sequence
 from scipy.ndimage import zoom
 from skimage.io import imread
-from tensorflow.keras.utils import Sequence
 
 from cellfinder.core import types
 from cellfinder.core.classify.augment import AugmentationParameters, augment
@@ -56,7 +56,14 @@ def __init__(
         translate: Tuple[float, float, float] = (0.05, 0.05, 0.05),
         shuffle: bool = False,
         interpolation_order: int = 2,
+        *args,
+        **kwargs,
     ):
+        # pass any additional arguments not specified in signature to the
+        # constructor of the superclass (e.g.: `use_multiprocessing` or
+        # `workers`)
+        super().__init__(*args, **kwargs)
+
         self.points = points
         self.signal_array = signal_array
         self.background_array = background_array
@@ -218,10 +225,10 @@ def __getitem__(self, index: int) -> Union[
 
         if self.train:
             batch_labels = [cell.type - 1 for cell in cell_batch]
-            batch_labels = tf.keras.utils.to_categorical(
+            batch_labels = keras.utils.to_categorical(
                 batch_labels, num_classes=self.classes
             )
-            return images, batch_labels
+            return images, batch_labels.astype(np.float32)
         elif self.extract:
             batch_info = self.__get_batch_dict(cell_batch)
             return images, batch_info
@@ -252,7 +259,8 @@ def __generate_cubes(
                 (number_images,)
                 + (self.cube_height, self.cube_width, self.cube_depth)
                 + (self.channels,)
-            )
+            ),
+            dtype=np.float32,
         )
 
         for idx, cell in enumerate(cell_batch):
@@ -350,7 +358,14 @@ def __init__(
         translate: Tuple[float, float, float] = (0.2, 0.2, 0.2),
         train: bool = False,  # also return labels
         interpolation_order: int = 2,
+        *args,
+        **kwargs,
     ):
+        # pass any additional arguments not specified in signature to the
+        # constructor of the superclass (e.g.: `use_multiprocessing` or
+        # `workers`)
+        super().__init__(*args, **kwargs)
+
         self.im_shape = shape
         self.batch_size = batch_size
         self.labels = labels
@@ -410,10 +425,10 @@ def __getitem__(self, index: int) -> Union[
 
         if self.train and self.labels is not None:
             batch_labels = [self.labels[k] for k in indexes]
-            batch_labels = tf.keras.utils.to_categorical(
+            batch_labels = keras.utils.to_categorical(
                 batch_labels, num_classes=self.classes
             )
-            return images, batch_labels
+            return images, batch_labels.astype(np.float32)
         else:
             return images
 
@@ -424,7 +439,8 @@ def __generate_cubes(
     ) -> np.ndarray:
         number_images = len(list_signal_tmp)
         images = np.empty(
-            ((number_images,) + self.im_shape + (self.channels,))
+            ((number_images,) + self.im_shape + (self.channels,)),
+            dtype=np.float32,
         )
 
         for idx, signal_im in enumerate(list_signal_tmp):
@@ -433,7 +449,7 @@ def __generate_cubes(
                 images, idx, signal_im, background_im
             )
 
-        return images.astype(np.float16)
+        return images
 
     def __populate_array_with_cubes(
         self,
diff --git a/cellfinder/core/classify/resnet.py b/cellfinder/core/classify/resnet.py
index ed313642..07f6253c 100644
--- a/cellfinder/core/classify/resnet.py
+++ b/cellfinder/core/classify/resnet.py
@@ -1,9 +1,11 @@
 from typing import Callable, Dict, List, Literal, Optional, Tuple, Union
 
-from tensorflow import Tensor
-from tensorflow.keras import Model
-from tensorflow.keras.initializers import Initializer
-from tensorflow.keras.layers import (
+from keras import (
+    KerasTensor as Tensor,
+)
+from keras import Model
+from keras.initializers import Initializer
+from keras.layers import (
     Activation,
     Add,
     BatchNormalization,
@@ -14,7 +16,7 @@
     MaxPooling3D,
     ZeroPadding3D,
 )
-from tensorflow.keras.optimizers import Adam, Optimizer
+from keras.optimizers import Adam, Optimizer
 
 #####################################################################
 # Define the types of ResNet
@@ -113,7 +115,7 @@ def non_residual_block(
     activation: str = "relu",
     use_bias: bool = False,
     bn_epsilon: float = 1e-5,
-    pooling_padding: str = "same",
+    pooling_padding: str = "valid",
     axis: int = 3,
 ) -> Tensor:
     """
@@ -131,6 +133,7 @@ def non_residual_block(
     )(x)
     x = BatchNormalization(axis=axis, epsilon=bn_epsilon, name="conv1_bn")(x)
     x = Activation(activation, name="conv1_activation")(x)
+
     x = MaxPooling3D(
         max_pool_size,
         strides=strides,
diff --git a/cellfinder/core/classify/tools.py b/cellfinder/core/classify/tools.py
index 2d5c44b2..3bf48876 100644
--- a/cellfinder/core/classify/tools.py
+++ b/cellfinder/core/classify/tools.py
@@ -1,9 +1,10 @@
 import os
-from typing import List, Optional, Sequence, Tuple, Union
+from collections.abc import Sequence
+from typing import List, Optional, Tuple, Union
 
+import keras
 import numpy as np
-import tensorflow as tf
-from tensorflow.keras import Model
+from keras import Model
 
 from cellfinder.core import logger
 from cellfinder.core.classify.resnet import build_model, layer_type
@@ -17,8 +18,7 @@ def get_model(
     inference: bool = False,
     continue_training: bool = False,
 ) -> Model:
-    """
-    Returns the correct model based on the arguments passed
+    """Returns the correct model based on the arguments passed
     :param existing_model: An existing, trained model. This is returned if it
     exists
     :param model_weights: This file is used to set the model weights if it
@@ -30,29 +30,31 @@ def get_model(
     by using the default one
     :param continue_training: If True, will ensure that a trained model
     exists. E.g. by using the default one
-    :return: A tf.keras model
+    :return: A keras model
 
     """
     if existing_model is not None or network_depth is None:
         logger.debug(f"Loading model: {existing_model}")
-        return tf.keras.models.load_model(existing_model)
+        return keras.models.load_model(existing_model)
     else:
         logger.debug(f"Creating a new instance of model: {network_depth}")
         model = build_model(
-            network_depth=network_depth, learning_rate=learning_rate
+            network_depth=network_depth,
+            learning_rate=learning_rate,
         )
         if inference or continue_training:
             logger.debug(
-                f"Setting model weights according to: {model_weights}"
+                f"Setting model weights according to: {model_weights}",
             )
             if model_weights is None:
-                raise IOError("`model_weights` must be provided")
+                raise OSError("`model_weights` must be provided")
             model.load_weights(model_weights)
         return model
 
 
 def make_lists(
-    tiff_files: Sequence, train: bool = True
+    tiff_files: Sequence,
+    train: bool = True,
 ) -> Union[Tuple[List, List], Tuple[List, List, np.ndarray]]:
     signal_list = []
     background_list = []
diff --git a/cellfinder/core/main.py b/cellfinder/core/main.py
index 926fe545..23526a94 100644
--- a/cellfinder/core/main.py
+++ b/cellfinder/core/main.py
@@ -1,23 +1,13 @@
-"""
-N.B imports are within functions to prevent tensorflow being imported before
-it's warnings are silenced
-"""
-
 import os
 from typing import Callable, List, Optional, Tuple
 
 import numpy as np
 from brainglobe_utils.cells.cells import Cell
-from brainglobe_utils.general.logging import suppress_specific_logs
 
 from cellfinder.core import logger
 from cellfinder.core.download.download import model_type
 from cellfinder.core.train.train_yml import depth_type
 
-tf_suppress_log_messages = [
-    "multiprocessing can interact badly with TensorFlow"
-]
-
 
 def main(
     signal_array: np.ndarray,
@@ -58,13 +48,11 @@ def main(
         Called every time a plane has finished being processed during the
         detection stage. Called with the plane number that has finished.
     classify_callback : Callable[int], optional
-        Called every time tensorflow has finished classifying a point.
+        Called every time a point has finished being classified.
         Called with the batch number that has just finished.
     detect_finished_callback : Callable[list], optional
         Called after detection is finished with the list of detected points.
     """
-    suppress_tf_logging(tf_suppress_log_messages)
-
     from cellfinder.core.classify import classify
     from cellfinder.core.detect import detect
     from cellfinder.core.tools import prep
@@ -98,7 +86,7 @@ def main(
     if not skip_classification:
         install_path = None
         model_weights = prep.prep_model_weights(
-            model_weights, install_path, model, n_free_cpus
+            model_weights, install_path, model
         )
         if len(points) > 0:
             logger.info("Running classification")
@@ -120,17 +108,4 @@ def main(
             )
         else:
             logger.info("No candidates, skipping classification")
-
     return points
-
-
-def suppress_tf_logging(tf_suppress_log_messages: List[str]) -> None:
-    """
-    Prevents many lines of logs such as:
-    "2019-10-24 16:54:41.363978: I tensorflow/stream_executor/platform/default
-    /dso_loader.cc:44] Successfully opened dynamic library libcuda.so.1"
-    """
-    os.environ["TF_CPP_MIN_LOG_LEVEL"] = "3"
-
-    for message in tf_suppress_log_messages:
-        suppress_specific_logs("tensorflow", message)
diff --git a/cellfinder/core/tools/prep.py b/cellfinder/core/tools/prep.py
index a2625311..6609d711 100644
--- a/cellfinder/core/tools/prep.py
+++ b/cellfinder/core/tools/prep.py
@@ -9,9 +9,7 @@
 from typing import Optional
 
 from brainglobe_utils.general.config import get_config_obj
-from brainglobe_utils.general.system import get_num_processes
 
-import cellfinder.core.tools.tf as tf_tools
 from cellfinder.core import logger
 from cellfinder.core.download.download import (
     DEFAULT_DOWNLOAD_DIRECTORY,
@@ -26,20 +24,13 @@ def prep_model_weights(
     model_weights: Optional[os.PathLike],
     install_path: Optional[os.PathLike],
     model_name: model_type,
-    n_free_cpus: int,
 ) -> Path:
-    n_processes = get_num_processes(min_free_cpu_cores=n_free_cpus)
-    prep_tensorflow(n_processes)
+    # prepare models (get default weights or provided ones)
     model_weights = prep_models(model_weights, install_path, model_name)
 
     return model_weights
 
 
-def prep_tensorflow(max_threads: int) -> None:
-    tf_tools.set_tf_threads(max_threads)
-    tf_tools.allow_gpu_memory_growth()
-
-
 def prep_models(
     model_weights_path: Optional[os.PathLike],
     install_path: Optional[os.PathLike],
diff --git a/cellfinder/core/tools/tf.py b/cellfinder/core/tools/tf.py
deleted file mode 100644
index 778aa78f..00000000
--- a/cellfinder/core/tools/tf.py
+++ /dev/null
@@ -1,46 +0,0 @@
-import tensorflow as tf
-
-from cellfinder.core import logger
-
-
-def allow_gpu_memory_growth():
-    """
-    If a gpu is present, prevent tensorflow from using all the memory straight
-    away. Allows multiple processes to use the GPU (and avoid occasional
-    errors on some systems) at the cost of a slight performance penalty.
-    """
-    gpus = tf.config.experimental.list_physical_devices("GPU")
-    if gpus:
-        logger.debug("Allowing GPU memory growth")
-        try:
-            # Currently, memory growth needs to be the same across GPUs
-            for gpu in gpus:
-                tf.config.experimental.set_memory_growth(gpu, True)
-            logical_gpus = tf.config.experimental.list_logical_devices("GPU")
-            logger.debug(
-                f"{len(gpus)} physical GPUs, {len(logical_gpus)} logical GPUs"
-            )
-        except RuntimeError as e:
-            # Memory growth must be set before GPUs have been initialized
-            print(e)
-    else:
-        logger.debug("No GPUs found, using CPU.")
-
-
-def set_tf_threads(max_threads):
-    """
-    Limit the number of threads that tensorflow uses
-    :param max_threads: Maximum number of threads to use
-    :return:
-    """
-    logger.debug(
-        f"Setting maximum number of threads for tensorflow "
-        f"to: {max_threads}"
-    )
-
-    # If statements are for testing. If tf is initialised, then setting these
-    # parameters throws an error
-    if tf.config.threading.get_inter_op_parallelism_threads() != 0:
-        tf.config.threading.set_inter_op_parallelism_threads(max_threads)
-    if tf.config.threading.get_intra_op_parallelism_threads() != 0:
-        tf.config.threading.set_intra_op_parallelism_threads(max_threads)
diff --git a/cellfinder/core/train/train_yml.py b/cellfinder/core/train/train_yml.py
index bf916b3c..4d6e5bf3 100644
--- a/cellfinder/core/train/train_yml.py
+++ b/cellfinder/core/train/train_yml.py
@@ -22,7 +22,10 @@
     check_positive_float,
     check_positive_int,
 )
-from brainglobe_utils.general.system import ensure_directory_exists
+from brainglobe_utils.general.system import (
+    ensure_directory_exists,
+    get_num_processes,
+)
 from brainglobe_utils.IO.cells import find_relevant_tiffs
 from brainglobe_utils.IO.yaml import read_yaml_section
 from fancylog import fancylog
@@ -33,11 +36,6 @@
 from cellfinder.core.classify.resnet import layer_type
 from cellfinder.core.download.download import DEFAULT_DOWNLOAD_DIRECTORY
 
-tf_suppress_log_messages = [
-    "sample_weight modes were coerced from",
-    "multiprocessing can interact badly with TensorFlow",
-]
-
 depth_type = Literal["18", "34", "50", "101", "152"]
 
 models: Dict[depth_type, layer_type] = {
@@ -318,11 +316,7 @@ def run(
     save_progress=False,
     epochs=100,
 ):
-    from cellfinder.core.main import suppress_tf_logging
-
-    suppress_tf_logging(tf_suppress_log_messages)
-
-    from tensorflow.keras.callbacks import (
+    from keras.callbacks import (
         CSVLogger,
         ModelCheckpoint,
         TensorBoard,
@@ -339,7 +333,6 @@ def run(
         model_weights=model_weights,
         install_path=install_path,
         model_name=model,
-        n_free_cpus=n_free_cpus,
     )
 
     yaml_contents = parse_yaml(yaml_file)
@@ -361,6 +354,7 @@ def run(
 
     signal_train, background_train, labels_train = make_lists(tiff_files)
 
+    n_processes = get_num_processes(min_free_cpu_cores=n_free_cpus)
     if test_fraction > 0:
         logger.info("Splitting data into training and validation datasets")
         (
@@ -387,15 +381,17 @@ def run(
             labels=labels_test,
             batch_size=batch_size,
             train=True,
+            use_multiprocessing=False,
+            workers=n_processes,
         )
 
         # for saving checkpoints
-        base_checkpoint_file_name = "-epoch.{epoch:02d}-loss-{val_loss:.3f}.h5"
+        base_checkpoint_file_name = "-epoch.{epoch:02d}-loss-{val_loss:.3f}"
 
     else:
         logger.info("No validation data selected.")
         validation_generator = None
-        base_checkpoint_file_name = "-epoch.{epoch:02d}.h5"
+        base_checkpoint_file_name = "-epoch.{epoch:02d}"
 
     training_generator = CubeGeneratorFromDisk(
         signal_train,
@@ -405,6 +401,8 @@ def run(
         shuffle=True,
         train=True,
         augment=not no_augment,
+        use_multiprocessing=False,
+        workers=n_processes,
     )
     callbacks = []
 
@@ -421,9 +419,14 @@ def run(
 
     if not no_save_checkpoints:
         if save_weights:
-            filepath = str(output_dir / ("weight" + base_checkpoint_file_name))
+            filepath = str(
+                output_dir
+                / ("weight" + base_checkpoint_file_name + ".weights.h5")
+            )
         else:
-            filepath = str(output_dir / ("model" + base_checkpoint_file_name))
+            filepath = str(
+                output_dir / ("model" + base_checkpoint_file_name + ".keras")
+            )
 
         checkpoints = ModelCheckpoint(
             filepath,
@@ -432,25 +435,26 @@ def run(
         callbacks.append(checkpoints)
 
     if save_progress:
-        filepath = str(output_dir / "training.csv")
-        csv_logger = CSVLogger(filepath)
+        csv_filepath = str(output_dir / "training.csv")
+        csv_logger = CSVLogger(csv_filepath)
         callbacks.append(csv_logger)
 
     logger.info("Beginning training.")
+    # Keras 3.0: `use_multiprocessing` input is set in the
+    # `training_generator` (False by default)
     model.fit(
         training_generator,
         validation_data=validation_generator,
-        use_multiprocessing=False,
         epochs=epochs,
         callbacks=callbacks,
     )
 
     if save_weights:
         logger.info("Saving model weights")
-        model.save_weights(str(output_dir / "model_weights.h5"))
+        model.save_weights(output_dir / "model.weights.h5")
     else:
         logger.info("Saving model")
-        model.save(output_dir / "model.h5")
+        model.save(output_dir / "model.keras")
 
     logger.info(
         "Finished training, " "Total time taken: %s",
diff --git a/pyproject.toml b/pyproject.toml
index fa717476..d31bda7e 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -29,9 +29,8 @@ dependencies = [
     "numpy",
     "scikit-image",
     "scikit-learn",
-    # See https://github.com/brainglobe/cellfinder-core/issues/103 for < 2.12.0 pin
-    "tensorflow-macos>=2.5.0,<2.12.0; platform_system=='Darwin' and platform_machine=='arm64'",
-    "tensorflow>=2.5.0,<2.12.0; platform_system!='Darwin' or platform_machine!='arm64'",
+    "keras>=3.0.0",
+    "torch>=2.1.0",
     "tifffile",
     "tqdm",
 ]
@@ -79,7 +78,7 @@ requires = ["setuptools>=45", "wheel", "setuptools_scm[toml]>=6.2"]
 build-backend = 'setuptools.build_meta'
 
 [tool.black]
-target-version = ['py39', 'py310']
+target-version = ['py39', 'py310','py311']
 skip-string-normalization = false
 line-length = 79
 
@@ -111,13 +110,14 @@ markers = ["slow: marks tests as slow (deselect with '-m \"not slow\"')"]
 legacy_tox_ini = """
 # For more information about tox, see https://tox.readthedocs.io/en/latest/
 [tox]
-envlist = py{39,310}
+envlist = py{39,310,311}
 isolated_build = true
 
 [gh-actions]
 python =
     3.9: py39
     3.10: py310
+    3.11: py311
 
 [testenv]
 commands = python -m pytest -v --color=yes --cov=cellfinder --cov-report=xml
@@ -132,6 +132,8 @@ deps =
     pytest-qt
 extras =
     napari
+setenv =
+    KERAS_BACKEND = torch
 passenv =
     NUMBA_DISABLE_JIT
     CI
diff --git a/tests/core/conftest.py b/tests/core/conftest.py
index 29e465ad..e97b878f 100644
--- a/tests/core/conftest.py
+++ b/tests/core/conftest.py
@@ -1,8 +1,10 @@
 import os
 from typing import Tuple
 
+import keras.src.backend.common.global_state
 import numpy as np
 import pytest
+import torch.backends.mps
 from skimage.filters import gaussian
 
 from cellfinder.core.download.download import (
@@ -11,6 +13,22 @@
 )
 
 
+@pytest.fixture(scope="session", autouse=True)
+def set_device_arm_macos_ci():
+    """
+    Ensure that the device is set to CPU when running on arm based macOS
+    GitHub runners. This is to avoid the following error:
+    https://discuss.pytorch.org/t/mps-back-end-out-of-memory-on-github-action/189773/5
+    """
+    if (
+        os.getenv("GITHUB_ACTIONS") == "true"
+        and torch.backends.mps.is_available()
+    ):
+        keras.src.backend.common.global_state.set_global_attribute(
+            "torch_device", "cpu"
+        )
+
+
 @pytest.fixture(scope="session")
 def no_free_cpus() -> int:
     """
diff --git a/tests/core/test_integration/test_detection.py b/tests/core/test_integration/test_detection.py
index 7d90a007..fc7bf2f3 100644
--- a/tests/core/test_integration/test_detection.py
+++ b/tests/core/test_integration/test_detection.py
@@ -80,7 +80,10 @@ def test_detection_full(signal_array, background_array, free_cpus, request):
 
 
 def test_detection_small_planes(
-    signal_array, background_array, no_free_cpus, mocker
+    signal_array,
+    background_array,
+    no_free_cpus,
+    mocker,
 ):
     # Check that processing works when number of planes < number of processes
     nproc = get_num_processes(no_free_cpus)
diff --git a/tests/core/test_integration/test_train.py b/tests/core/test_integration/test_train.py
index 0d4f5c38..8cfc8c31 100644
--- a/tests/core/test_integration/test_train.py
+++ b/tests/core/test_integration/test_train.py
@@ -35,5 +35,5 @@ def test_train(tmpdir):
     sys.argv = train_args
     train_run()
 
-    model_file = os.path.join(tmpdir, "model.h5")
+    model_file = os.path.join(tmpdir, "model.keras")
     assert os.path.exists(model_file)