DLLA: Dynamic Learnable Label Assignment for Indoor 3D Object Detection

This repository contains an implementation of DLLA, a dynamiclearnable label assignment for indoor 3D Object Detection method introduced in our paper:

Dynamic Learnable Label Assignment for Indoor 3D Object Detection
Xinrun Liu, Linqin Zhao, Bin Fan, Jiwen Lu, Hongmin Liu
University of Science and Technology Beijing, Tsinghua University

Installation

You can install all required packages manually. This implementation is based on mmdetection3d framework. Please refer to the original installation guide getting_started.md, including MinkowskiEngine installation, replacing open-mmlab/mmdetection3d with liuxinrun/dlla.

Most of the DLLA-related code locates in the following files: detectors/mink_single_stage.py, dense_heads/dlla_tr3d_head.py, dense_heads/dlla_fcaf3d_head.py,

Getting Started

Please see getting_started.md for basic usage examples. We follow the mmdetection3d data preparation protocol described in scannet, sunrgbd.

Training

To start training, run train with DLLA for TR3D configs:

python tools/train.py configs/dlla/dlla_tr3d_scannet-3d-18class.py

Run train with DLLA for FCAF3D configs:

python tools/train.py configs/dlla/dlla_fcaf3d_scannet-3d-18class.py

Testing

Test pre-trained model using test with DLLA for TR3D configs:

python tools/test.py configs/dlla/dlla_tr3d_scannet-3d-18class.py \
    work_dirs/dlla_tr3d_scannet-3d-18class/latest.pth --eval mAP

Visualization

Visualizations can be created with test script. For better visualizations, you may set score_thr in configs to 0.3:

python tools/test.py configs/dlla/dlla_tr3d_scannet-3d-18class.py \
    work_dirs/dlla_tr3d_scannet-3d-18class/latest.pth --eval mAP --show \
    --show-dir work_dirs/dlla_tr3d_scannet-3d-18class

Models

The metrics are obtained in 15 training runs followed by 15 test runs. We report both the best and the average values (the latter are given in round brackets).

DLLA 3D Detection

Model	mAP@0.25	mAP@0.5	Download
DLLA_TR3D	73.8 (72.8)	60.2 (58.9)	model | log | config
DLLA_FCAF3D	71.4 (71.0)	60.0 (59.0)	model | log | config