SOLQ: Segmenting Objects by Learning Queries

This repository is an official implementation of the NeurIPS 2021 paper SOLQ: Segmenting Objects by Learning Queries.

Introduction

TL; DR. SOLQ is an end-to-end instance segmentation framework with Transformer. It directly outputs the instance masks without any box dependency.

Abstract. In this paper, we propose an end-to-end framework for instance segmentation. Based on the recently introduced DETR, our method, termed SOLQ, segments objects by learning unified queries. In SOLQ, each query represents one object and has multiple representations: class, location and mask. The object queries learned perform classification, box regression and mask encoding simultaneously in an unified vector form. During training phase, the mask vectors encoded are supervised by the compression coding of raw spatial masks. In inference time, mask vectors produced can be directly transformed to spatial masks by the inverse process of compression coding. Experimental results show that SOLQ can achieve state-of-the-art performance, surpassing most of existing approaches. Moreover, the joint learning of unified query representation can greatly improve the detection performance of original DETR. We hope our SOLQ can serve as a strong baseline for the Transformer-based instance segmentation.

Updates

• (12/10/2021) Release D-DETR+SQR log.txt in SQR.
• (29/09/2021) Our SOLQ has been accepted by NeurIPS 2021.
• (14/07/2021) Higher performance (Box AP=56.5, Mask AP=46.7) is reported by training with long side 1536 on Swin-L backbone, instead of long side 1333.

Main Results

Installation

The codebase is built on top of Deformable DETR.

Requirements

• Linux, CUDA>=9.2, GCC>=5.4

• Python>=3.7

  We recommend you to use Anaconda to create a conda environment:

  conda create -n deformable_detr python=3.7 pip

  Then, activate the environment:

  conda activate deformable_detr

• PyTorch>=1.5.1, torchvision>=0.6.1 (following instructions here)

  For example, if your CUDA version is 9.2, you could install pytorch and torchvision as following:

  conda install pytorch=1.5.1 torchvision=0.6.1 cudatoolkit=9.2 -c pytorch

• Other requirements

  pip install -r requirements.txt

• Build MultiScaleDeformableAttention

  cd ./models/ops
  sh ./make.sh

Usage

Dataset preparation

Please download COCO and organize them as following:

mkdir data && cd data
ln -s /path/to/coco coco

Training and Evaluation

Training on single node

Training SOLQ on 8 GPUs as following:

sh configs/r50_solq_train.sh

Evaluation

You can download the pretrained model of SOLQ (the link is in "Main Results" session), then run following command to evaluate it on COCO 2017 val dataset:

sh configs/r50_solq_eval.sh

Evaluation on COCO 2017 test-dev dataset

You can download the pretrained model of SOLQ (the link is in "Main Results" session), then run following command to evaluate it on COCO 2017 test-dev dataset (submit to server):

sh configs/r50_solq_submit.sh

Visualization on COCO 2017 val dataset

You can visualize on image as follows:

EXP_DIR=/path/to/checkpoint
python visual.py \
       --meta_arch solq \
       --backbone resnet50 \
       --with_vector \
       --with_box_refine \
       --masks \
       --batch_size 2 \
       --vector_hidden_dim 1024 \
       --vector_loss_coef 3 \
       --output_dir ${EXP_DIR} \
       --resume ${EXP_DIR}/solq_r50_final.pth \
       --eval    

Citing SOLQ

If you find SOLQ useful in your research, please consider citing:

@article{dong2021solq,
  title={SOLQ: Segmenting Objects by Learning Queries},
  author={Dong, Bin and Zeng, Fangao and Wang, Tiancai and Zhang, Xiangyu and Wei, Yichen},
  journal={NeurIPS},
  year={2021}
}