NeRI

Official Implementation of "NERI: IMPLICIT NEURAL REPRESENTATION OF LIDAR POINT CLOUD USING RANGE IMAGE SEQUENCE"

Get Started

Data Preparation

The dataset used in this work can be downloaded from KITTI. It should be organized as follows:

• sequence/00/
  • velodyne/
    • 000000.bin
    • 000001.bin
    • 000002.bin
    • ......
  • calib.txt
  • pose.txt
• ......

Dependencies

You can install the dependencies using the following command：

conda create -n neri
conda activate neri
pip install -r requirements.txt 

Training

Here is an example command for training:

python train.py -e 600 --outf tp --stem_dim_num 64_1 --fc_hw_dim 4_125_26  --single_res --act swish --eval_freq=1 --temporal_embed='1.25_20' --translation_embed='1.25_30' --rotation_embed='1.25_30'  --segmentation --cfg='config/kitti_00.yaml' --strides 2 2 2 2

Testing

Here is an example command for testing:

python train.py -e 600 --outf tp --stem_dim_num 64_1 --fc_hw_dim 4_125_26  --single_res --act swish --eval_freq=1 --temporal_embed='1.25_20' --translation_embed='1.25_30' --rotation_embed='1.25_30'  --segmentation --cfg='config/kitti_00.yaml' --strides 2 2 2 2 --eval_only --quant_bit=-1

python train.py -e 600 --outf tp --stem_dim_num 64_1 --fc_hw_dim 4_125_26  --single_res --act swish --eval_freq=1 --temporal_embed='1.25_20' --translation_embed='1.25_30' --rotation_embed='1.25_30'   --segmentation --cfg='config/kitti_00.yaml' --strides 2 2 2 2 --eval_only --quant_mode='pw-1' --quant_bit=16

The bpp (bits per point) can be calculated by:

$$Bpp = ModelSize \times QuantizationPrecision \times EntropyEfficiency / TotalNumPoints$$

Contact

If you have any questions, concerns, or would like to contribute to this project, feel free to reach out. You can contact us at：

• Email: [email protected]