d-fine-medium

<h1 align="center"><strong>D-FINE Medium</strong></h1> <p align="center"> <a href="https://huggingface.co/Laudando-Associates-LLC/d-fine-medium"> <img src="https://img.shields.io/badge/HuggingFace-Model-yellow?logo=huggingface&style=for-the-badge"> </a> </p>

This repository contains the D-FINE Medium model, a real-time object detector designed for efficient and accurate object detection tasks.

<p align="center"> <img src="assets/medium.png" alt="Medium Detections" /> </p>

Try it in the Browser

You can test this model using our interactive Gradio demo:

<p align="center"> <a href="https://huggingface.co/spaces/Laudando-Associates-LLC/d-fine-demo"> <img src="https://img.shields.io/badge/Launch%20Demo-Gradio-FF4B4B?logo=gradio&logoColor=white&style=for-the-badge"> </a> </p>

Model Overview

• Architecture: D-FINE Medium

• Parameters: 19.6M

• Performance:

  • mAP@[0.50:0.95]: 0.840

  • mAP@[0.50]: 0.992

  • AR@[0.50:0.95]: 0.894

  • F1 Score: 0.924

• Framework: PyTorch / ONNX

• Training Hardware: 2× NVIDIA RTX A6000 GPUs

Download

Usage

To utilize this model, ensure you have the shared D-FINE processor:

from transformers import AutoProcessor, AutoModel

# Load processor
processor = AutoProcessor.from_pretrained("Laudando-Associates-LLC/d-fine", trust_remote_code=True)

# Load model
model = AutoModel.from_pretrained("Laudando-Associates-LLC/d-fine-medium", trust_remote_code=True)

# Process image
inputs = processor(image)

# Run inference
outputs = model(**inputs, conf_threshold=0.4)

Evaluation

This model was trained and evaluated on the L&A Pucks Dataset.

License

This model is licensed under the Apache License 2.0.

Citation

If you use D-FINE or its methods in your work, please cite the following BibTeX entries:

@misc{peng2024dfine,
      title={D-FINE: Redefine Regression Task in DETRs as Fine-grained Distribution Refinement},
      author={Yansong Peng and Hebei Li and Peixi Wu and Yueyi Zhang and Xiaoyan Sun and Feng Wu},
      year={2024},
      eprint={2410.13842},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}