Single Stage Common Object Detection: A Simple Baseline

Official Pytorch implementation for the paper "Single Stage Class Agnostic Common Object Detection". This work is based on MMDetection 1.1.0.

Installation

conda env create -f environment.yml -n sscod
conda activate sscod

pip install "git+https://github.com/cocodataset/cocoapi.git#subdirectory=PythonAPI"
pip install -v -e .

Training and Testing

Scripts for training and testing models are put in folder scripts:

scripts
├── coco
│   ├── test
│   │   ├── exp2_caseA_baseline.sh
│   │   ├── exp2_caseB_baseline.sh
│   │   └── exp2_caseB_curcon.sh
│   └── train
│       ├── exp2_caseA_baseline.sh
│       ├── exp2_caseB_baseline.sh
│       └── exp2_caseB_curcon.sh
└── voc
    ├── test
    │   ├── exp2_arcconneg.sh
    │   ├── exp2_arccon.sh
    │   ├── exp2_baseline.sh
    │   ├── exp2_curcon.sh
    │   └── exp2_focalcur.sh
    └── train
        ├── exp2_arcconneg.sh
        ├── exp2_arccon.sh
        ├── exp2_baseline.sh
        ├── exp2_curcon.sh
        └── exp2_focalcur.sh

Checkpoints can be downloaded from https://www.dropbox.com/sh/j151jxfz44dzl2i/AACSTybiL5u9nC20x0aSzwR3a?dl=0.

Quantitative results

VOC dataset

COCO dataset

Qualitative results