/radiant

Primary LanguagePython

RADIANT: Radar-Image Association Network for 3D Object Detection

example figure Visualization of predicted radar offsets (cyan arrows) to object centers and detections on image and BEV. The radar association of the RADIANT corrects the localization error of the monocular methods, improving detection performance. Orange, magenta and cyan boxes are GT bounding boxes, monocular detections, and RADIANT detections, respectively.

Directories

radiant/
    data/                           							 
        nuscenes/                 		    
                maps/
                samples/
                sweeps/
                v1.0-trainval/
                v1.0-test/
                fusion_data/                
    lib/
    scripts/
    checkpoints/

Setup

  • Create a conda environment
conda create -n radiant python=3.6

  • Install required packages (see requirements.txt)

  • Download nuScenes dataset (full dataset (v1.0)) into data/nuscenes/

Code

The code of monocular components, i.e. FCOS3D and PGD, is adapted from OpenMMLab

Data Preparation

cd radiant
python scripts/prepare_data_trainval.py
python scripts/prepare_data_test.py

FCOS3D + RADIANT

1. Train radar branch

Download weights of FCOS3D pretrained with nuScenes training or training + validation set

CUDA_VISIBLE_DEVICES=0,1,2,3 python scripts/train_radiant_fcos3d.py --resume --num_gpus 4 --samples_per_gpu 4 --epochs 12 --lr 0.001 --workers_per_gpu 2

2. Train depth weight net (DWN)

# 1) Generate training data
CUDA_VISIBLE_DEVICES=0 python scripts/prepare_data_dwn_fcos3d.py 

# 2) Train
CUDA_VISIBLE_DEVICES=0,1,2,3 python scripts/train_dwn.py --resume --workers_per_gpu 2 --samples_per_gpu 256 --num_gpus 4 --epochs 200

3. Evaluation

On validation set

Download pretrained weights of camera/radar branch and DWN

# Evalutate fusion method FCOS3D + RADIANT
CUDA_VISIBLE_DEVICES=0 python scripts/train_radiant_fcos3d.py --do_eval 

# Evalutate monocular method FCOS3D
CUDA_VISIBLE_DEVICES=0 python scripts/train_radiant_fcos3d.py --do_eval --eval_mono
On test set

Download pretrained weights of camera/radar branch and DWN

# Evalutate fusion method FCOS3D + RADIANT
CUDA_VISIBLE_DEVICES=0 python scripts/train_radiant_fcos3d.py \
--do_eval --eval_set test \
--dir_result data/nuscenes/fusion_data/train_result/radiant_fcos3d_test \
--path_checkpoint_dwn data/nuscenes/fusion_data/dwn_radiant_fcos3d_test/train_result/checkpoint.tar 

# Evalutate monocular method FCOS3D
CUDA_VISIBLE_DEVICES=0 python scripts/train_radiant_fcos3d.py \
--do_eval --eval_mono --eval_set test \
--dir_result data/nuscenes/fusion_data/train_result/radiant_fcos3d_test \
--path_checkpoint_dwn data/nuscenes/fusion_data/dwn_radiant_fcos3d_test/train_result/checkpoint.tar

PGD + RADIANT

1. Train radar branch

Download weights of PGD pretrained with nuScenes training set

CUDA_VISIBLE_DEVICES=0,1,2,3 python scripts/train_radiant_pgd.py --resume --num_gpus 4 --samples_per_gpu 4 --epochs 10 --lr 0.001 --workers_per_gpu 2

2. Train DWN

# 1) Generate training data
CUDA_VISIBLE_DEVICES=0 python scripts/prepare_data_dwn_pgd.py 

# 2) Train
CUDA_VISIBLE_DEVICES=0,1,2,3 python scripts/train_dwn.py --resume --workers_per_gpu 2 --samples_per_gpu 256 --num_gpus 4 --epochs 200
--dir_data data/nuscenes/fusion_data/dwn_radiant_pgd

3. Evaluation

Download pretrained weights of camera/radar branch and DWN

On validation set
# Evalutate fusion method PGD + RADIANT
CUDA_VISIBLE_DEVICES=0 python scripts/train_radiant_pgd.py --do_eval 

# Evalutate monocular method PGD
CUDA_VISIBLE_DEVICES=0 python scripts/train_radiant_pgd.py --do_eval --eval_mono
On test set
# Evalutate fusion method PGD + RADIANT
CUDA_VISIBLE_DEVICES=0 python scripts/train_radiant_pgd.py --do_eval --eval_set test

# Evalutate monocular method PGD
CUDA_VISIBLE_DEVICES=0 python scripts/train_radiant_pgd.py --do_eval --eval_mono --eval_set test

Citation

@InProceedings{long2023radiant,
    author    = {Long, Yunfei and Kumar, Abhinav and Morris, Daniel and Liu, Xiaoming and Castro, Marcos and Chakravarty, Punarjay},
    title     = {RADIANT: Radar-Image Association Network for 3D Object Detection},
    booktitle = {AAAI},
    year      = {2023}
}