Name		Name	Last commit message	Last commit date
Latest commit History 4 Commits
.gitattributes		.gitattributes
.gitignore		.gitignore
Dataloader_loss.py		Dataloader_loss.py
README.md		README.md
boston_singapore.py		boston_singapore.py
data_helper.py		data_helper.py
dataset-nuscenes.sh		dataset-nuscenes.sh
helper.py		helper.py
main.py		main.py
model.py		model.py
nuscenes.txt		nuscenes.txt
nuscenes_data_process.ipynb		nuscenes_data_process.ipynb
nuscenes_tutorial.ipynb		nuscenes_tutorial.ipynb
params.py		params.py
pose_visualizer.py		pose_visualizer.py
preprocess.py		preprocess.py
test.py		test.py
visualize.py		visualize.py

Repository files navigation

DeepVO-pytorch

Tools/Methods: PyTorch, FlowNetSimple, LSTM, MSE, Weights & Biases

(1) Utilized rotation matrix and geometric equation to acquire the absolute and relative poses of various images

(2) Reproduced DeepVO network architectures by integrating FlowNetSimple pretrained model with LSTM

(3) Compiled loss function by using Mean Square Error (MSE) of all positions $p$ and orientations $\varphi$

(4) Evaluated the influence of epochs and hyperparameter $k$ in the loss function on the DeepVO network using KITTI and nuScenes datasets according to translation and rotation RMSE values

(5) Plotted trajectories on the predicted results from test sequences and compared them with original routes

About

DeepVO model for nuSences datasets

Report repository

Releases

No releases published

Packages

No packages published

Languages