FARP-Net: Local-Global Feature Aggregation and Relation-Aware Proposals for 3D Object Detection

This is a MMDetection3D implementation of the paper "FARP-Net: Local-Global Feature Aggregation and Relation-Aware Proposals for 3D Object Detection".

Prerequisites

The code is tested with Python3.7, PyTorch == 1.10, CUDA == 11.3, mmdet3d == 1.0.0rc2, mmcv_full == 1.5.0 and mmdet == 2.24.1. We recommend you to use anaconda to make sure that all dependencies are in place. Note that different versions of the library may cause changes in results.

Step 1. Create a conda environment and activate it.

conda create --name pt1.10.v1 python=3.7
conda activate pt1.10.v1

Step 2. Install MMDetection3D following the instruction here.

Step 3. Prepare SUN RGB-D Data following the procedure here.

Getting Started

for sunrgbd

sh tools/slurm_train.sh $PARTION $JOB_NAME configs/A2FRPG/A2FRPG_16x8_sunrgbd-3d-10class.py $WORK_DIR

for scannet-1x-backbone

sh tools/slurm_train.sh $PARTION $JOB_NAME configs/configs/A2FRPG/A2FRPG_8x8_scannet-3d-18class.py $WORK_DIR

for scannet-2x-backbone

sh tools/slurm_train.sh $PARTION $JOB_NAME configs/configs/A2FRPG/A2FRPG_8x8_scannet-3d-18class-2x.py $WORK_DIR

for test the pretrained weight

sh tools/slurm_test.sh $PARTION $JOB_NAME configs/A2FRPG/A2FRPG_16x8_sunrgbd-3d-10class.py $PRETRAINED_CKPT --eval mAP --work-dir $WORK_DIR

Main Results

SUNRGB-D

name	Lr schd	mAP@0.25	Download
A2FRPGNet	3x	64.1	model \| log

ScanNet

name	Lr schd	backbone	mAP@0.25	Download
A2FRPGNet	3x	1x	69.1	model \| log
A2FRPGNet	3x	2x	70.9	model \| log

yeminglang/FARP-Net