Deep Visual Odometry

Training and validation of Visual Odometry models. Implementation is based on with a few changes.

Running with wandb

Create a new project at https://app.wandb.ai/

WANDB_NAME='testing' CUDA_VISIBLE_DEVICES=2 python3 train.py config_kitti.yaml

Here:

WANDB_NAME - Experiment name
CUDA_VISIBLE_DEVICES - ID of a GPU used for training (starting with 0) To see which GPUs are free: nvidia-smi
config_kitti.yaml - Parameters of the experiment and path to the dataset.

Name		Name	Last commit message	Last commit date
Latest commit History 2 Commits
data		data
EDA.ipynb		EDA.ipynb
README.md		README.md
_t265_convert_poses.py		_t265_convert_poses.py
config_d455.yaml		config_d455.yaml
config_kitti.yaml		config_kitti.yaml
config_kitti_transfer.yaml		config_kitti_transfer.yaml
config_t265.yaml		config_t265.yaml
data_utils.py		data_utils.py
geometry_utils.py		geometry_utils.py
kitti_odometry.py		kitti_odometry.py
load_config.py		load_config.py
losses_and_functions.py		losses_and_functions.py
make_film.py		make_film.py
models.py		models.py
plot_utils.py		plot_utils.py
requirements.txt		requirements.txt
t265_debug.ipynb		t265_debug.ipynb
test_onnx.py		test_onnx.py
train.py		train.py
train_eval_utils.py		train_eval_utils.py