car-studio

Project Page | Paper | Data

News:

[2024/02/26] Our codes have been released, feel free to explore it!
[2024/01/02] Car-Studio is accepted to RA-L!

Install

conda create -n car-studio python=3.8 pip -y
conda activate car-studio

git submodule update --init
cd dependencies/nerfstudio
pip install -e .

cd ../../dependencies/segment-anything
pip install -e .

cd ../../
pip install -e .

pip uninstall nvidia-cublas-cu11
pip install nvidia-cublas-cu12
pip install git+https://github.com/NVlabs/tiny-cuda-nn/#subdirectory=bindings/torch

Preparation of the dataset

KITTI Multi-Object Tracking(MOT)

To prepare the KITTI multi-object tracking dataset, follow these steps:

Download KITTI multi-oject tracking dataset from https://www.cvlibs.net/datasets/kitti/eval_tracking.php
- Download left color images of tracking data set (15 GB);
- Download camera calibration matrices of tracking data set (1 MB);
- Download training labels of tracking data set (9 MB).
Unzip the downloaded files and place them in the directory ./data/kitti-mot.

Process KITTI MOT

Strongly recommended, you can directly download the processed dataset 'km.zip' from the following link: here

Alternatively, you could also process them manually, but need to dive into the code and have some modifications. To convert the original KITTI multi-object tracking dataset to the CarPatch3D dataset, follow these steps:

Download the pre-trained model for segment-anything:

wget https://dl.fbaipublicfiles.com/segment_anything/sam_vit_h_4b8939.pth -O dependencies/segment-anything/sam_vit_h_4b8939.pth

Run the script to process the KITTI MOT dataset:

python car_studio/scripts/datasets/process_kitti.py --dataset km --data_dir "./data/kitti-mot"

Process More Data(Optional)

Using the same method as for the KITTI MOT dataset, you can also process or download the KITTI OBJ dataset and the DVM-CAR dataset. Please note that processing or downloading all three datasets simultaneously is not necessary. Using the KITTI MOT dataset alone can achieve an out-of-the-box performance.

If you want to process the DVM CARS dataset, you will need additional dependencies:

wget https://tri-ml-public.s3.amazonaws.com/github/dd3d/experiments/4elbgev2-20210825_201852/model_final.pth

pip install hydra fvcore
pip install 'git+https://github.com/facebookresearch/detectron2.git'

pip install "git+https://github.com/facebookresearch/pytorch3d.git@stable"

pip install mpi4py
pip install seaborn numba

File Tree of the Dataset

Ensure that the directory tree of the dataset is organized as follows, with the car_studio and kitti-mot folders in the same directory:

./data/
|-- car_studio
|   |-- dv.json
|   |-- km.json
|   |-- ko.json
|   |-- mask
|   `-- patch
|-- kitti-mot
|   |-- testing
|   `-- training
|-- kitti-obj
|   |-- testing
|   `-- training
`-- dvm
    `-- resized_DVM

Train

Our code is based on the code architecture of nerfstudio. We recommend considering referring to the nerfstudio tutorial to enhance your understanding of our code and to assist with any configuration or modifications you may wish to make.

Codes to train a car-nerf:

cd car_studio
ns-train car-nerf

Test

Codes to test a car-nerf, (Please make sure to replace <path> in load_dir with the appropriate path to your checkpoint path.):

cd car_studio
ns-train car-nerf --max_num_iterations 10 --load_dir <outputs/.../nerfstudio_models/>  --steps_per_eval_all_
images 1

Bibtex

If this work is helpful for your research, please cite the following BibTeX entry.

@misc{liu2023carstudio,
      title={Car-Studio: Learning Car Radiance Fields from Single-View and Endless In-the-wild Images}, 
      author={Tianyu Liu and Hao Zhao and Yang Yu and Guyue Zhou and Ming Liu},
      year={2023},
      eprint={2307.14009},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

Acknowledgment

We extend our gratitude to the exceptional open-source projects:

lty2226262/Car_studio