---
license: apache-2.0
library_name: pytorch
tags:
  - object-detection
  - computer-vision
  - discrete-diffusion
  - wacv-2027
datasets:
  - detection-datasets/coco
---

# DDM-Det: Fast Object Detection via One-Step Discrete Diffusion

Official pretrained model checkpoint for **Fast Object Detection via One-Step Discrete Diffusion**, accepted at **WACV 2027**.

DDM-Det formulates object detection as discrete token generation and performs detection in a single reverse diffusion step.

## Results

Evaluated on COCO 2017 validation using a ResNet-50 backbone.

| AP | AP50 | AP75 | FPS |
|---|---|---|---|
| 40.50 | 60.58 | 42.29 | 58.82 |

*FPS is the paper-reported end-to-end inference speed, measured under the experimental setup described in the paper.*

## Model Checkpoint

- **File:** `model_final.pth`
- **Backbone:** ResNet-50
- **Dataset:** COCO 2017
- **Framework:** PyTorch and Detectron2

## Code and Usage

The complete implementation, installation instructions, and evaluation commands are available on GitHub:

**[DDM-Det GitHub Repository](https://github.com/neeresh/one-step-discrete-diffusion)**

## Citation

If you use this model in your research, please cite:

*Fast Object Detection via One-Step Discrete Diffusion.* IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), 2027.

The official publication and BibTeX citation will be added when available.

## License

Apache License 2.0.