DLU implementation in Pytorch

Code for the paper:

[Lighten CARAFE: Dynamic Lightweight Upsampling with Guided Reassemble Kernels]

Oral at ICPR 2024,
Authors:
Ruigang Fu, Qingyong Hu, Xiaohu Dong, Yinghui Gao, Biao Li and Ping Zhong

---

Requirements

1. Environment:

The requirements are exactly the same as mmdetection(https://github.com/open-mmlab/mmdetection). We tested on the following settings:

• python 3.8
• cuda 10.1
• pytorch 1.8.1+cu101
• torchvision 0.9.1+cu101
• mmcv 2.1.0

conda create -n dlu python=3.8 -y
source activate dlu
conda activate dlu
pip install torch==1.8.1+cu101 torchvision==0.9.1+cu101 torchaudio==0.8.1 -f https://download.pytorch.org/whl/torch_stable.html
pip install -U openmim
mim install mmengine
mim install "mmcv>=2.0.0"
git clone https://github.com/Fu0511/Dynamic-Lightweight-Upsampling.git
cd mmdetection
pip install -v -e .

2. Data:

The folder data should be like this:

    data
    ├── coco
    │   ├── annotations
    │   ├── train2017
    │   ├── val2017
    │   ├── test2017

Training

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}

Config files in our experiments are under ./custom/config/FPN/.

• ./custom/config/FPN/coco_custom_dlu.py: FPN model with DLU as its upsampling operation.
• ./custom/config/FPN/coco_custom_carafe: FPN model with carafe as its upsampling operation.
• ./custom/config/FPN/coco_baseline_bilinear: FPN model with bilinear as its upsampling operation.

Results

The main results on coco val set:

Method	AP	Additional Model Size
Bilinear	37.5	--
CARAFE	38.6	+1.2MB
DLU	38.6	+0.6MB

---

If these codes are useful to you, please cite our work:

@misc{fu2024Lighten,
    title={Lighten CARAFE: Dynamic Lightweight Upsampling with Guided Reassemble Kernels},
    author={Ruigang Fu and Qingyong Hu and Xiaohu Dong and Yinghui Gao and Biao Li and Ping Zhong},
    year={2024},
    eprint={2410.22139},
    archivePrefix={arXiv},
    primaryClass={cs.CV}
}