Inference Installation instructions
Please use Python 2.7. Setup a conda environment, install matplotlib, pytorch (tested with 0.4.1), OpenCV 3.0 and our library lbd_mod, including the python interface, for line detection.
Download archive and unzip it to the repo folder.
Run python infer.py
Train Installation instructions
Please use Python 2.7. Setup a conda environment, install matplotlib, pytorch (tested with 0.4.1), OpenCV 3.0 and our library lbd_mod, including the python interface, for line detection.
Download LLD Dataset and unzip it to the '../batched' folder with respect to the cloned repo location.
Download KITTI Odometry and unzip it to the '../kitti' folder with respect to the cloned repo location.
Download EuRoC MAV zip files and unzip it to the '../euroc' folder with respect to the cloned repo location.
Download the EuRoC camera parameters file EuRoC YAML and save it as '../EuRoC_Empty.yaml'.
Rectify EuRoC dataset using python rectify_euroc.py ../euroc ../euroc_rect ../EuRoC_Empty.yaml
Create a folder ../kittieuroc
Combine EuRoC and KITTI using python prepare_kitti_euroc_combined.py ../kitti ../euroc_rect ../kittieuroc
Run python train.py. It calls train_multibatch to train the network, eval_multibatch to evaluate and test_with_descriptors_hetero
to save the descriptors to use with the LLD-SLAM.
Please cite:
A. Vakhitov, V. Lempitsky, Learnable Line Descriptor for Visual Navigation, IEEE Access, 2019