ang-ang/saic_t5d2

use pointpillars as baseline, use SAIC dataset to do 3D detection

SAIC_T5D2

Use pointpillars as baseline, use SAIC dataset to do 3D detection

Forked from: https://github.com/nutonomy/second.pytorch

Pre_trained model download:

Model_car

Model_pedestrian

Install

1. Clone code

git clone https://github.com/Ang-ang/SAIC_T5D2.git

2. Install Python packages

Use Anaconda to configure as many packages as possible.

conda create -n SAIC python=3.7 anaconda
source activate SAIC
conda install numpy=1.17.2 shapely pybind11 protobuf scikit-image numba pillow
conda install pytorch torchvision cudatoolkit=10.1 -c pytorch

3. Setup cuda for numba

Add following environment variables for numba to ~/.bashrc:

export NUMBAPRO_CUDA_DRIVER=/usr/lib/x86_64-linux-gnu/libcuda.so
export NUMBAPRO_NVVM=/usr/local/cuda/nvvm/lib64/libnvvm.so
export NUMBAPRO_LIBDEVICE=/usr/local/cuda/nvvm/libdevice

4. PYTHONPATH

Add SAIC_T5D2/ to PYTHONPATH.

Prepare dataset

1. Dataset preparation

Download SAIC dataset and create some directories first:

└── SAIC_DATASET_ROOT
       ├── training
       |   ├── calibration
       |   ├── label
       |   ├── velodyne
       └── testing
           ├── velodyne

Note: this repo use SAIC_DATASET_ROOT=/data/SAIC_dataset/.

2. Create infos:

python create_data.py create_kitti_info_file --data_path=SAIC_DATASET_ROOT

3. Create groundtruth-database infos:

python create_data.py create_groundtruth_database --data_path=SAIC_DATASET_ROOT

4. Modify config file

The config file needs to be edited to point to the above datasets:

train_input_reader: {
  ...
  database_sampler {
    database_info_path: "/path/to/SAIC_dataset_dbinfos_train.pkl"
    ...
  }
  kitti_info_path: "/path/to/SAIC_dataset_infos_train.pkl"
  kitti_root_path: "SAIC_DATASET_ROOT"
}
...
eval_input_reader: {
  ...
  kitti_info_path: "/path/to/SAIC_dataset_infos_val.pkl"
  kitti_root_path: "SAIC_DATASET_ROOT"
}

Train

cd ~/second.pytorch/second/pytorch
python train.py train --config_path=(path to config file) --model_dir=(path to model dir)

Evaluate

cd ~/second.pytorch/second/pytorch
python train.py evaluate --config_path=(path to config file) --model_dir=(path to model dir) --ckpt_path=(path to model)