diff --git a/bin/preprocess_data.py b/bin/preprocess_data.py
new file mode 100644
index 0000000..4aaec48
--- /dev/null
+++ b/bin/preprocess_data.py
@@ -0,0 +1,80 @@
+#!/usr/bin/env python3
+# Copyright 2021 by Julian Nubert, Robotic Systems Lab, ETH Zurich.
+# All rights reserved.
+# This file is released under the "BSD-3-Clause License".
+# Please see the LICENSE file that has been included as part of this package.
+import os
+
+import numpy as np
+import torch
+import yaml
+
+import preprocessing.preprocesser
+
+
+def yes_or_no(question):
+    while "the answer is invalid":
+        reply = str(input(question + ' (y/n): ')).lower().strip()
+        if reply[0] == 'y':
+            return True
+        else:
+            return False
+
+
+def config():
+    # Load parameters
+    f = open('config/config_datasets.yaml')
+    config = yaml.load(f, Loader=yaml.FullLoader)
+    f = open('config/deployment_options.yaml')
+    deployment_options = yaml.load(f, Loader=yaml.FullLoader)
+    config.update(deployment_options)
+
+    # Device to be used
+    config["device"] = torch.device(config["device"])
+
+    for dataset in config["datasets"]:
+        config[dataset]["horizontal_cells"] = config[dataset]["horizontal_cells_preprocessing"]
+        config[dataset]["data_identifiers"] = config[dataset]["training_identifiers"] + config[dataset][
+            "testing_identifiers"]
+
+    # Convert angles to radians
+    for dataset in config["datasets"]:
+        config[dataset]["vertical_field_of_view"][0] *= (np.pi / 180.0)
+        config[dataset]["vertical_field_of_view"][1] *= (np.pi / 180.0)
+    config["horizontal_field_of_view"][0] *= (np.pi / 180.0)
+    config["horizontal_field_of_view"][1] *= (np.pi / 180.0)
+
+    # Check whether rosbag exists
+    for dataset in config["datasets"]:
+        print("Checking whether path to " + config[dataset]["data_path"] + " exists.")
+        if not os.path.exists(config[dataset]["data_path"]):
+            raise Exception("Path " + config[dataset]["data_path"] + " does not exist. Exiting.")
+
+    # User check for correctness of paths -------------
+    print("----------------------------------")
+    print("Run for the datasets: " + str(config["datasets"]))
+    print("which are located at")
+    for dataset in config["datasets"]:
+        print(config[dataset]["data_path"])
+    print("and will be stored at")
+    for dataset in config["datasets"]:
+        print(config[dataset]["preprocessed_path"])
+    print("----------")
+    if not yes_or_no("Continue?"):
+        print("Okay, then program will be stopped.")
+        exit()
+
+    # -------------------------------------------------
+
+    print("----------------------------------")
+    print("Configuration for this run: ")
+    print(config)
+    print("----------------------------------")
+
+    return config
+
+
+if __name__ == "__main__":
+    config = config()
+    preprocesser = preprocessing.preprocesser.Preprocesser(config=config)
+    preprocesser.preprocess_data()
diff --git a/bin/run_rosnode.py b/bin/run_rosnode.py
new file mode 100644
index 0000000..7a561ee
--- /dev/null
+++ b/bin/run_rosnode.py
@@ -0,0 +1,77 @@
+#!/usr/bin/env python3
+# Copyright 2021 by Julian Nubert, Robotic Systems Lab, ETH Zurich.
+# All rights reserved.
+# This file is released under the "BSD-3-Clause License".
+# Please see the LICENSE file that has been included as part of this package.
+import click
+import numpy as np
+import torch
+import yaml
+
+import ros_utils.odometry_publisher
+
+
+@click.command()
+@click.option('--checkpoint', prompt='Path to the saved model you want to test')
+@click.option('--dataset',
+              prompt='On which dataset configuration do you want to get predictions? [kitti, darpa, ....]. Does not '
+                     'need to be one of those, but the sensor paramaters are looked up in the config_datasets.yaml.')
+@click.option('--lidar_topic', prompt='Topic of the published LiDAR pointcloud2 messages.')
+@click.option('--lidar_frame', prompt='LiDAR frame in TF tree.')
+@click.option('--integrate_odometry', help='Whether the published odometry should be integrated in the TF tree.',
+              default=True)
+def config(checkpoint, dataset, lidar_topic, lidar_frame, integrate_odometry):
+    f = open('config/config_datasets.yaml')
+    config = yaml.load(f, Loader=yaml.FullLoader)
+    f = open('config/deployment_options.yaml')
+    deployment_options = yaml.load(f, Loader=yaml.FullLoader)
+    config.update(deployment_options)
+    f = open('config/hyperparameters.yaml')
+    network_hyperparameters = yaml.load(f, Loader=yaml.FullLoader)
+    config.update(network_hyperparameters)
+
+    # Mode
+    config["mode"] = "training"
+
+    # No dropout during testing
+    if config["use_dropout"]:
+        config["use_dropout"] = False
+        print("Deactivating dropout for this mode.")
+
+    # CLI Input
+    ## Checkpoint
+    config["checkpoint"] = str(checkpoint)
+    ## Dataset
+    config["datasets"] = [str(dataset)]
+    ## LiDAR Topic
+    config["lidar_topic"] = str(lidar_topic)
+    ## LiDAR Frame
+    config["lidar_frame"] = str(lidar_frame)
+    ## Integrate odometry
+    config["integrate_odometry"] = integrate_odometry
+
+    # Device to be used
+    if config["device"] == "cuda":
+        config["device"] = torch.device("cuda")
+    else:
+        config["device"] = torch.device("cpu")
+
+    # Convert angles to radians
+    for dataset in config["datasets"]:
+        config[dataset]["vertical_field_of_view"][0] *= (np.pi / 180.0)
+        config[dataset]["vertical_field_of_view"][1] *= (np.pi / 180.0)
+    config["horizontal_field_of_view"][0] *= (np.pi / 180.0)
+    config["horizontal_field_of_view"][1] *= (np.pi / 180.0)
+
+    print("----------------------------------")
+    print("Configuration for this run: ")
+    print(config)
+    print("----------------------------------")
+
+    return config
+
+
+if __name__ == "__main__":
+    config = config(standalone_mode=False)
+    publisher = ros_utils.odometry_publisher.OdometryPublisher(config=config)
+    publisher.publish_odometry()
diff --git a/bin/run_testing.py b/bin/run_testing.py
new file mode 100644
index 0000000..0799a68
--- /dev/null
+++ b/bin/run_testing.py
@@ -0,0 +1,97 @@
+#!/usr/bin/env python3
+# Copyright 2021 by Julian Nubert, Robotic Systems Lab, ETH Zurich.
+# All rights reserved.
+# This file is released under the "BSD-3-Clause License".
+# Please see the LICENSE file that has been included as part of this package.
+import click
+import numpy as np
+import torch
+import yaml
+
+import deploy.tester
+
+
+@click.command()
+@click.option('--testing_run_name', prompt='MLFlow name of the run',
+              help='The name under which the run can be found afterwards.')
+@click.option('--experiment_name', help='High-level testing sequence name for clustering in MLFlow.',
+              default="testing")
+@click.option('--checkpoint', prompt='Path to the saved checkpoint of the model you want to test')
+def config(testing_run_name, experiment_name, checkpoint):
+    f = open('config/config_datasets.yaml')
+    config = yaml.load(f, Loader=yaml.FullLoader)
+    f = open('config/deployment_options.yaml')
+    deployment_options = yaml.load(f, Loader=yaml.FullLoader)
+    config.update(deployment_options)
+    f = open('config/hyperparameters.yaml')
+    network_hyperparameters = yaml.load(f, Loader=yaml.FullLoader)
+    config.update(network_hyperparameters)
+
+    # Parameters from previous run?
+    if 'parameters' in torch.load(checkpoint):
+        print("\033[92m" +
+              "Found parameters in checkpoint! Setting part of parameters to those ones."
+              + "\033[0;0m")
+        parameters_exist = True
+    else:
+        print("Checkpoint does not contain any parameters. Using those ones specified in the YAML files.")
+        parameters_exist = False
+
+    # Parameters that are set depending on whether provided in checkpoint
+    if parameters_exist:
+        loaded_config = torch.load(checkpoint)['parameters']
+        ## Device to be used
+        loaded_config["device"] = torch.device(config["device"])
+        ## Dataset selection
+        loaded_config["datasets"] = config["datasets"]
+        for dataset in loaded_config["datasets"]:
+            loaded_config[dataset]["testing_identifiers"] = config[dataset]["testing_identifiers"]
+            loaded_config[dataset]["data_identifiers"] = loaded_config[dataset]["testing_identifiers"]
+        ## Inference only
+        loaded_config["inference_only"] = config["inference_only"]
+        loaded_config["store_dataset_in_RAM"] = config["store_dataset_in_RAM"]
+        config = loaded_config
+    # Some parameters are only initialized when not taken from checkpoint
+    else:
+        ## Device to be used
+        config["device"] = torch.device(config["device"])
+        for dataset in config["datasets"]:
+            config[dataset]["data_identifiers"] = config[dataset]["testing_identifiers"]
+        ## Convert angles to radians
+        for dataset in config["datasets"]:
+            config[dataset]["vertical_field_of_view"][0] *= (np.pi / 180.0)
+            config[dataset]["vertical_field_of_view"][1] *= (np.pi / 180.0)
+        config["horizontal_field_of_view"][0] *= (np.pi / 180.0)
+        config["horizontal_field_of_view"][1] *= (np.pi / 180.0)
+
+    # Parameters that are always set
+    ## No dropout during testing
+    if config["use_dropout"]:
+        config["use_dropout"] = False
+        print("Deactivating dropout for this mode.")
+
+    ## CLI Input
+    ### Testing run name
+    config["run_name"] = str(testing_run_name)
+    ### Checkpoint
+    config["checkpoint"] = str(checkpoint)
+    ### Experiment name, default specified in deployment_options.yaml
+    if experiment_name:
+        config["experiment"] = experiment_name
+    ## Mode
+    config["mode"] = "testing"
+    ## Unsupervised
+    config["unsupervised_at_start"] = True
+
+    print("----------------------------------")
+    print("Configuration for this run: ")
+    print(config)
+    print("----------------------------------")
+
+    return config
+
+
+if __name__ == "__main__":
+    config = config(standalone_mode=False)
+    tester = deploy.tester.Tester(config=config)
+    tester.test()
diff --git a/bin/run_training.py b/bin/run_training.py
new file mode 100644
index 0000000..380c211
--- /dev/null
+++ b/bin/run_training.py
@@ -0,0 +1,94 @@
+#!/usr/bin/env python3
+# Copyright 2021 by Julian Nubert, Robotic Systems Lab, ETH Zurich.
+# All rights reserved.
+# This file is released under the "BSD-3-Clause License".
+# Please see the LICENSE file that has been included as part of this package.
+import click
+import numpy as np
+import torch
+import yaml
+
+import deploy.trainer
+
+
+@click.command()
+@click.option('--training_run_name', prompt='MLFlow name of the run',
+              help='The name under which the run can be found afterwards.')
+@click.option('--experiment_name', help='High-level training sequence name for clustering in MLFlow.',
+              default="")
+@click.option('--checkpoint', help='Path to the saved checkpoint. Leave empty if none.',
+              default="")
+def config(training_run_name, experiment_name, checkpoint):
+    f = open('config/config_datasets.yaml')
+    config = yaml.load(f, Loader=yaml.FullLoader)
+    f = open('config/deployment_options.yaml')
+    deployment_options = yaml.load(f, Loader=yaml.FullLoader)
+    config.update(deployment_options)
+    f = open('config/hyperparameters.yaml')
+    network_hyperparameters = yaml.load(f, Loader=yaml.FullLoader)
+    config.update(network_hyperparameters)
+
+    # Default: load parameters from yaml
+    parameters_exist = False
+
+    # CLI Input
+    ## Checkpoint for continuing training
+    if checkpoint:
+        ### Parameters from previous run?
+        if 'parameters' in torch.load(checkpoint):
+            print("\033[92m" +
+                  "Found parameters in checkpoint of previous run! Setting part of parameters to those ones."
+                  + "\033[0;0m")
+            parameters_exist = True
+        else:
+            print("Checkpoint does not contain any parameters. Using those ones specified in the YAML files.")
+
+    # Parameters that are set depending on whether provided in checkpoint
+    if parameters_exist:
+        loaded_config = torch.load(checkpoint)['parameters']
+        ## Device to be used
+        loaded_config["device"] = torch.device(config["device"])
+        loaded_config["datasets"] = config["datasets"]
+        for dataset in loaded_config["datasets"]:
+            loaded_config[dataset]["training_identifiers"] = config[dataset]["training_identifiers"]
+            loaded_config[dataset]["data_identifiers"] = loaded_config[dataset]["training_identifiers"]
+        config = loaded_config
+    # Some parameters are only initialized when not taken from checkpoint
+    else:
+        ## Device to be used
+        config["device"] = torch.device(config["device"])
+        for dataset in config["datasets"]:
+            config[dataset]["data_identifiers"] = config[dataset]["training_identifiers"]
+        ## Convert angles to radians
+        for dataset in config["datasets"]:
+            config[dataset]["vertical_field_of_view"][0] *= (np.pi / 180.0)
+            config[dataset]["vertical_field_of_view"][1] *= (np.pi / 180.0)
+        config["horizontal_field_of_view"][0] *= (np.pi / 180.0)
+        config["horizontal_field_of_view"][1] *= (np.pi / 180.0)
+
+    # Parameters that are always set
+    if checkpoint:
+        config["checkpoint"] = str(checkpoint)
+    else:
+        config["checkpoint"] = None
+    ## Trainings run name --> mandatory
+    config["training_run_name"] = str(training_run_name)
+    config["run_name"] = config["training_run_name"]
+    ## Experiment name, default specified in deployment_options.yaml
+    if experiment_name:
+        config["experiment"] = experiment_name
+    ## Mode
+    config["mode"] = "training"
+
+    print("----------------------------------")
+    print("Configuration for this run: ")
+    print(config)
+    print("----------------------------------")
+
+    return config
+
+
+if __name__ == "__main__":
+    config = config(standalone_mode=False)
+    trainer = deploy.trainer.Trainer(config=config)
+    trainer.train()
diff --git a/bin/visualize_pointcloud_normals.py b/bin/visualize_pointcloud_normals.py
new file mode 100644
index 0000000..f08e443
--- /dev/null
+++ b/bin/visualize_pointcloud_normals.py
@@ -0,0 +1,51 @@
+#!/usr/bin/env python3
+# Copyright 2021 by Julian Nubert, Robotic Systems Lab, ETH Zurich.
+# All rights reserved.
+# This file is released under the "BSD-3-Clause License".
+# Please see the LICENSE file that has been included as part of this package.
+import numpy as np
+import torch
+import yaml
+
+import ros_utils.publish_point_cloud_and_normals
+
+def config():
+    f = open('config/config_datasets.yaml')
+    config = yaml.load(f, Loader=yaml.FullLoader)
+    f = open('config/deployment_options.yaml')
+    deployment_options = yaml.load(f, Loader=yaml.FullLoader)
+    config.update(deployment_options)
+
+    # Device to be used
+    if config["device"] == "cuda":
+        config["device"] = torch.device("cuda")
+    else:
+        config["device"] = torch.device("cpu")
+
+    # Data identifiers
+    for dataset in config["datasets"]:
+        if config["mode"] == "training":
+            config[dataset]["data_identifiers"] = config[dataset]["training_identifiers"]
+        elif config["mode"] == "testing":
+            config[dataset]["data_identifiers"] = config[dataset]["testing_identifiers"]
+        else:
+            raise Exception('Only modes "training" and "testing" are valid.')
+
+    # Convert angles to radians
+    for dataset in config["datasets"]:
+        config[dataset]["vertical_field_of_view"][0] *= (np.pi / 180.0)
+        config[dataset]["vertical_field_of_view"][1] *= (np.pi / 180.0)
+    config["horizontal_field_of_view"][0] *= (np.pi / 180.0)
+    config["horizontal_field_of_view"][1] *= (np.pi / 180.0)
+
+    print("----------------------------------")
+    print("Configuration for this run: ")
+    print(config)
+    print("----------------------------------")
+
+    return config
+
+if __name__ == "__main__":
+    config = config()
+    publisher = ros_utils.publish_point_cloud_and_normals.ROSPublisher(config=config)
+    publisher.publish_dataset()
\ No newline at end of file
diff --git a/scripts/convert_kitti_to_rosbag.py b/scripts/convert_kitti_to_rosbag.py
new file mode 100644
index 0000000..290b92f
--- /dev/null
+++ b/scripts/convert_kitti_to_rosbag.py
@@ -0,0 +1,48 @@
+#!/usr/bin/env python3
+# Copyright 2021 by Julian Nubert, Robotic Systems Lab, ETH Zurich.
+# All rights reserved.
+# This file is released under the "BSD-3-Clause License".
+# Please see the LICENSE file that has been included as part of this package.
+import numpy as np
+import torch
+import yaml
+
+import ros_utils.convert_to_rosbag
+
+
+def config():
+    f = open('config/deployment_options.yaml')
+    config = yaml.load(f, Loader=yaml.FullLoader)
+    f = open('config/config_datasets.yaml')
+    dataset_config = yaml.load(f, Loader=yaml.FullLoader)
+    config.update(dataset_config)
+
+    # Device to be used
+    if config["device"] == "cuda":
+        config["device"] = torch.device("cuda")
+    else:
+        config["device"] = torch.device("cpu")
+
+    for dataset in config["datasets"]:
+        config[dataset]["data_identifiers"] = config[dataset]["training_identifiers"] + config[dataset][
+            "testing_identifiers"]
+
+    # Convert angles to radians
+    for dataset in config["datasets"]:
+        config[dataset]["vertical_field_of_view"][0] *= (np.pi / 180.0)
+        config[dataset]["vertical_field_of_view"][1] *= (np.pi / 180.0)
+    config["horizontal_field_of_view"][0] *= (np.pi / 180.0)
+    config["horizontal_field_of_view"][1] *= (np.pi / 180.0)
+
+    print("----------------------------------")
+    print("Configuration for this run: ")
+    print(config)
+    print("----------------------------------")
+
+    return config
+
+
+if __name__ == "__main__":
+    config = config()
+    converter = ros_utils.convert_to_rosbag.RosbagConverter(config=config)
+    converter.convert()
diff --git a/scripts/convert_pytorch_models.py b/scripts/convert_pytorch_models.py
new file mode 100644
index 0000000..27feb3f
--- /dev/null
+++ b/scripts/convert_pytorch_models.py
@@ -0,0 +1,20 @@
+#!/usr/bin/env python3
+# Copyright 2021 by Julian Nubert, Robotic Systems Lab, ETH Zurich.
+# All rights reserved.
+# This file is released under the "BSD-3-Clause License".
+# Please see the LICENSE file that has been included as part of this package.
+import click
+import torch
+
+
+@click.command()
+@click.option('--checkpoint',
+              prompt='Path to the saved model (without .pth) you want to convert to older PyTorch compatibility.')
+def convert_pytorch_model(checkpoint):
+    state_dict = torch.load(checkpoint + ".pth", map_location=torch.device("cpu"))
+    print(state_dict)
+    torch.save(state_dict, checkpoint + "_py27.pth", _use_new_zipfile_serialization=False)
+
+
+if __name__ == "__main__":
+    convert_pytorch_model()
diff --git a/scripts/time_network.py b/scripts/time_network.py
new file mode 100644
index 0000000..9d4c867
--- /dev/null
+++ b/scripts/time_network.py
@@ -0,0 +1,93 @@
+#!/usr/bin/env python3
+# Copyright 2021 by Julian Nubert, Robotic Systems Lab, ETH Zurich.
+# All rights reserved.
+# This file is released under the "BSD-3-Clause License".
+# Please see the LICENSE file that has been included as part of this package.
+import click
+import numpy as np
+import torch
+import yaml
+
+import models.model
+import time
+
+from torch.utils import mkldnn as mkldnn_utils
+
+
+@click.command()
+@click.option('--checkpoint', help='Path to the saved model you want to continue training from.',
+              default="")
+def config(checkpoint):
+    f = open('config/deployment_options.yaml')
+    config = yaml.load(f, Loader=yaml.FullLoader)
+    f = open('config/config_datasets.yaml')
+    dataset_config = yaml.load(f, Loader=yaml.FullLoader)
+    config.update(dataset_config)
+    f = open('config/hyperparameters.yaml')
+    hyperparameters_config = yaml.load(f, Loader=yaml.FullLoader)
+    config.update(hyperparameters_config)
+
+    # CLI Input
+    if not checkpoint:
+        config["checkpoint"] = None
+    else:
+        config["checkpoint"] = str(checkpoint)
+
+    # Device to be used
+    config["device"] = torch.device(config["device"])
+
+    # Convert angles to radians
+    for dataset in config["datasets"]:
+        config[dataset]["vertical_field_of_view"][0] *= (np.pi / 180.0)
+        config[dataset]["vertical_field_of_view"][1] *= (np.pi / 180.0)
+    config["horizontal_field_of_view"][0] *= (np.pi / 180.0)
+    config["horizontal_field_of_view"][1] *= (np.pi / 180.0)
+
+    print("Configuration for this run: ")
+    print(config)
+
+    return config
+
+
+if __name__ == "__main__":
+    config = config(standalone_mode=False)
+    iterations = 1000
+    torch.set_num_threads(4)
+
+    # CUDA synchronisation
+    torch.cuda.synchronize()
+
+    # Velodyne VLP-16
+    print("Velodyne VLP-16 --------------")
+    sample_input = torch.rand(1, 4, 16, 720).to(config["device"])
+    print("Used device is: " + str(config["device"]))
+    ## Standard Model
+    model = models.model.OdometryModel(config=config).to(config["device"]).eval()
+    _, _ = model(sample_input, sample_input)
+    torch.cuda.synchronize()
+    t_accum = 0.0
+    for iteration in range(iterations):
+        t = time.time()
+        _, _ = model(sample_input, sample_input)
+        torch.cuda.synchronize()
+        t_delta = time.time() - t
+        t_accum += t_delta
+        print(str(t_delta * 1000) + "ms")
+    print("Average execution time of model is: " + str(t_accum / iterations * 1000) + " milliseconds.")
+
+    del model
+    model_jit = torch.jit.trace(
+        models.model.OdometryModel(config=config).to(config["device"]),
+        example_inputs=(sample_input, sample_input)).eval()
+    t_accum = 0.0
+    for iteration in range(iterations + 1):
+        torch.cuda.synchronize()
+        t = time.time()
+        _, _ = model_jit(sample_input, sample_input)
+        torch.cuda.synchronize()
+        t_delta = time.time() - t
+        if iteration != 0:
+            t_accum += t_delta
+        print(t_delta)
+    print(
+        "Average execution time of jit model is: " + str(t_accum / iterations * 1000) + " milliseconds.")