Awesome

Reconciling Object-Level and Global-Level Objectives for Long-Tail Detection

This repo is the official implementation of the ICCV 2023 paper Reconciling Object-Level and Global-Level Objectives for Long-Tail Detection by Shaoyu Zhang, Chen Chen, and Silong Peng.

framework

Requirements

Python 3.6+
PyTorch 1.8+
torchvision 0.9+
mmdet 2.25
mmcv 1.4

Usage

1. Install

# Clone the ROG repository.
git clone https://github.com/EricZsy/ROG.git
cd ROG 

# Create conda environment.
conda create --name rog python=3.8 -y 
conda activate rog
conda install pytorch torchvision torchaudio cudatoolkit

# Install mmcv and mmdetection.
pip install -U openmim
mim install mmcv-full==1.4.0
pip install mmdet==2.25.2 
pip install -v -e .

2. Data

Please download LVIS dataset. The folder data should be like this:

    data
    ├── lvis
    │   ├── lvis_annotations
    │   │   │   ├── lvis_v1_train.json
    │   │   │   ├── lvis_v1_val.json
    │   ├── train2017
    │   │   ├── 000000100582.jpg
    │   │   ├── 000000102411.jpg
    │   │   ├── ......
    │   └── val2017
    │       ├── 000000062808.jpg
    │       ├── 000000119038.jpg
    │       ├── ......

3. Train

Use the following commands to train a model.

# Single GPU
python tools/train.py ${CONFIG_FILE}

# Multi GPU distributed training
./tools/dist_train.sh ${CONFIG_FILE} ${GPU_NUM} [optional arguments]

For example, to train a Mask R-CNN model for 12 epochs with ROG:

# Single GPU
python tools/train.py configs/rog/rog_r50_sample1e-3_1x.py

# Multi GPU distributed training (for 4 gpus)
bash ./tools/dist_train.sh configs/rog/rog_r50_sample1e-3_1x.py 4

Other configs can be found at ./configs/rog/. You may also use custom loss or sampling method with ROG.

4. Test

Use the following commands to test a trained model.

bash ./tools/dist_test.sh \
 configs/rog/rog_r50_sample1e-3_1x.py work_dirs/rog_r50_sample1e-3_1x.py/latest.pth 4 --eval bbox segm

Citation

If you find this work useful in your research, please cite:

@inproceedings{zhang2023reconciling,
    title={Reconciling Object-Level and Global-Level Objectives for Long-Tail Detection},
    author={Zhang, Shaoyu and Chen, Chen and Peng, Silong},
    booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
    pages={18982--18992},
    year={2023}
}

Acknowledgement

Thanks MMDetection team for the wonderful open source project!