Neural Pose Estimation Based on Improved Detector-free Matching for Low-texture Environments

Neural Pose Estimation Based on Improved Detector-free Matching for Low-texture Environments

Installation

docker build -t dftransvo:1.0 .
docker run -it --user=root -v uploading_code/:/temp:rw --gpus --shm-size 32G –name build_env dftransvo:1.0
cd LoFTR
conda env create -f environment.yaml
conda activate dftransvo
pip install -r requirements.txt
exit
docker commit build_env dftransvo:2.0

Test

docker run –rm -it --user=root -v uploading_code/:/temp:rw --gpus --shm-size 32G –name test dftransvo:2.0
python test_pair.py

Acknowledgements

This project build upon the excellent work of:

• LoFTR
• EfficientLoFTR