Please check Slides and Project Report for more details and figures
Point Cloud Sequence with 3D BBox tracked and its Corresponding Video Sequence
This codebase is built on top of https://github.com/Ghostish/Open3DSOT, with references from https://github.com/HaozheQi/P2B and https://github.com/haooozi/OSP2B. Our Codebase was setup on GCP with 100gb storage for this project and NVIDIA L4 GPU(24GB RAM) and g2-standard-8 8vcpus 32GB RAM
cd 3dsot conda create -n myenv python=3.8 conda activate myenv
pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu117
pip install -r requirement.txt
--> [calib]
--> {0000-0020}.txt
--> [label_02]
--> {0000-0020}.txt
--> [velodyne]
--> [0000-0020] folders with velodynes .bin files
CUDA_VISIBLE_DEVICES=0 python main.py --cfg cfgs/P2B_Car.yaml --batch_size 64 --epoch 60 --category_name Car
python main.py --cfg cfgs/P2B_Car.yaml --checkpoint ./ckpts/car-31.ckpt --test --category_name Car
pip install tensorboard
tensorboard --logdir=./ --port=6006