3D Object Detection with SECOND on MORAI dataset

This repository is based on OpenPCDet. All configurations and codes were revised for MORAI dataset.

Results and Models

SECOND

Class accuracy is IoU(Intersection over Union).

Dataset	epoch	bbox AP(0.70)	bev AP(0.70)	bev AP(0.50)	3d AP(0.70)	3d AP(0.50)	aos	config	log	model
Real	80	84.46	62.83	74.31	50.25	73.52	82.70	config	[log]	[model]
Daegu	80	84.99	74.59	76.46	62.33	74.77	79.15	config	[log]	[model]
Daegu(Mix)	80	89.15	82.55	82.95	79.16	82.95	85.35	config	[log]	[model]
Sejong BRT 1	80	83.83	69.59	70.00	64.84	69.88	80.02	config	[log]	[model]
Sejong BRT 1(Mix)	80	81.18	70.19	70.45	61.56	70.21	76.22	config	[log]	[model]
Sangam Edge	80	81.56	66.73	69.28	63.02	68.00	65.41	config	[log]	[model]
Sangam Edge(Mix)	80	86.13	74.10	78.45	69.27	76.16	65.09	config	[log]	[model]
Sejong BRT 1 Edge	80	77.42	47.88	51.16	45.33	48.04	73.58	config	[log]	[model]
Sejong BRT 1 Edge(Mix)	80	88.93	80.99	81.34	80.39	81.31	85.76	config	[log]	[model]

Real:

Daegu:

Sejong BRT 1:

Sangam Edge:

Sejong BRT 1 Edge:

Usage

Installation

Please refer to install.md for installation, dataset preparation and making configuration file.

Testing

python test.py --cfg_file ${CONFIG_FILE} --batch_size ${BATCH_SIZE} --ckpt ${CKPT}

Example:

python test.py --cfg_file cfgs/custom_models/second.yaml --batch_size 4 --ckpt ../ckpt/custom_model/second_daegu.pth

Training

python train.py --cfg_file ${CONFIG_FILE}

Example:

python train.py --cfg_file cfgs/custom_models/second.yaml

Demo

python demo.py --cfg_file ${CONFIG_FILE} \
    --ckpt ${CKPT} \
    --data_path ${POINT_CLOUD_DATA}
    (--ext .npy)

Default --ext is .pcd. But custom point cloud is .npy

Example:

python demo.py --cfg_file cfgs/custom_models/second.yaml \
    --ckpt ../ckpt/custom_model/second_daegu.pth \
    --data_path ../data/custom/points/000000.npy
    --ext .npy