Hyper-YOLO: When Visual Object Detection Meets Hypergraph Computation.

Yifan Feng, Jiangang Huang, Shaoyi Du, Shihui Ying, Jun-Hai Yong, Yipeng Li, Guiguang Ding, Rongrong Ji, and Yue Gao.

Our Hyper-YOLO enhances the YOLOv8 architecture with hypergraph computation. The comparsion of four scale models are provided in the following table.

Furthermore, we replace the neck of YOLOv9 with the proposed HyperC2Net of our Hyper-YOLO, termed Hyper-YOLOv1.1. The source codes of Hyper-YOLOv1.1 are avaiable in here.

Clearly, in each scale, the Hyper-YOLOv1.1 outperforms the YOLOv9, which demonstrates the effectiveness of our HyperC2Net in capturing high-order feature correlations. The comparison of four scale models are provided in the following table

Clone repo and create conda environment (recommended). Then install requirements.txt in a Python>=3.8.0 environment, including PyTorch>=1.8. The command is as follows.

git clone https://github.com/iMoonLab/Hyper-YOLO.git  # clone
cd Hyper-YOLO
conda create -n Hyper-YOLO python=3.8
conda activate Hyper-YOLO
pip install -r requirements.txt  # install

You can also use the environment.yaml file and the conda command to install the required environment.

conda env create -f environment.yaml

Data Preparation: Download the MS COCO dataset images (training, validation, and test sets) and corresponding labels, or prepare your custom dataset as shown below. Additionally, modify the dataset path in ultralytics/cfg/datasets/coco.yaml to reflect the location of your data.

coco
--images
  --train2017
  --val2017
--labels
  --train2017
  --val2017

Most of training configurations can change in the "Train settings" section of ultralytics/cfg/default.yaml. The key factors are model, data, img, epoches, batch, device and training hyperparameters. For example, you can use "model: hyper-yolon.yaml" to train an object detection model.

python ultralytics/models/yolo/detect/train.py 

Most of evaluation configurations can change in the "Val/Test settings" section of ultralytics/cfg/default.yaml. The key factors are model(weight), data, img, batch, conf, iou, half.

python ultralytics/models/yolo/detect/val.py

Most of predict configurations can change in the "Predict settings" section of ultralytics/cfg/default.yaml. The key factors are model(weight), source, img, conf, iou.

python ultralytics/models/yolo/detect/predict.py

Here, our instance segmentation model configuration remains consistent with YOLOv8, by changing the object detection output head to an instance segmentation output head, while the backbone and neck remain unchanged. The model's training and testing process is similar to that of object detection. For example, you can use "model: hyper-yolon-seg.yaml" to train an instance segmentation model.

python ultralytics/models/yolo/segment/train.py
python ultralytics/models/yolo/segment/val.py
python ultralytics/models/yolo/segment/predict.py 

Here is an example code for exporting an ONNX model. If you need to export other formats, please refer to our example code and the YOLOv8 documentation.

python ultralytics/utils/export_onnx.py

Our code is built based on the YOLOv8. Thanks for their great work!

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

@article{feng2024hyper,
  title={Hyper-YOLO: When Visual Object Detection Meets Hypergraph Computation},
  author={Feng, Yifan and Huang, Jiangang and Du, Shaoyi and Ying, Shihui and Yong, Jun-Hai and Li, Yipeng and Ding, Guiguang and Ji, Rongrong and Gao, Yue},
  journal={arXiv preprint arXiv:2408.04804},
  year={2024}
}

Hypergraph computation is a powerful tool to capture high-order correlations among visual features. Compared with graphs, each hyperedge in a hypergraph can connect more than two vertices, which is more flexible to model complex correlations. Now, learning with high-order correlations still remains a under-explored area in computer vision. We hope our work can inspire more research in this direction. If you are interested in hypergraph computation, please refer to our series of works on hypergraph computation in the follows:

• Hypergraph Learning: Methods and Practices
• Hypergraph Nerual Networks
• HGNN+: General Hypergraph Nerual Networks
• Hypergraph Isomorphism Computation

Hyper-YOLO is maintained by iMoon-Lab, Tsinghua University. If you have any questions, please feel free to contact us via email: Yifan Feng and Jiangang Huang.