# RF-DETR: SOTA Real-Time Object Detection Model
[](https://badge.fury.io/py/rfdetr)
[](https://pypistats.org/packages/rfdetr)
[](https://badge.fury.io/py/rfdetr)
[](https://github.com/roboflow/rfdetr/blob/main/LICENSE)
[](https://huggingface.co/spaces/SkalskiP/RF-DETR)
[](https://colab.research.google.com/github/roboflow-ai/notebooks/blob/main/notebooks/how-to-finetune-rf-detr-on-detection-dataset.ipynb)
[](https://blog.roboflow.com/rf-detr)
[](https://discord.gg/GbfgXGJ8Bk)
RF-DETR is a real-time, transformer-based object detection model architecture developed by Roboflow and released under the Apache 2.0 license.
RF-DETR is the first real-time model to exceed 60 AP on the [Microsoft COCO benchmark](https://cocodataset.org/#home) alongside competitive performance at base sizes. It also achieves state-of-the-art performance on [RF100-VL](https://github.com/roboflow/rf100-vl), an object detection benchmark that measures model domain adaptability to real world problems. RF-DETR is fastest and most accurate for its size when compared current real-time objection models.
RF-DETR is small enough to run on the edge using [Inference](https://github.com/roboflow/inference), making it an ideal model for deployments that need both strong accuracy and real-time performance.
[Read the documentation to get started training.](https://rfdetr.roboflow.com)
## News
- `2025/07/23`: We release three new checkpoints for RF-DETR: Nano, Small, and Medium.
- RF-DETR Base is now deprecated. We recommend using RF-DETR Medium which offers subtantially better accuracy at comparable latency.
- `2025/03/20`: We release RF-DETR real-time object detection model. **Code and checkpoint for RF-DETR-large and RF-DETR-base are available.**
- `2025/04/03`: We release early stopping, gradient checkpointing, metrics saving, training resume, TensorBoard and W&B logging support.
- `2025/05/16`: We release an 'optimize_for_inference' method which speeds up native PyTorch by up to 2x, depending on platform.
## Results
RF-DETR achieves state-of-the-art performance on both the Microsoft COCO and the RF100-VL benchmarks.
The table below shows the performance of RF-DETR medium, compared to comparable medium models:
|family|size |coco_map50|coco_map50@95|rf100vl_map50|rv100vl_map50@95|latency|
|------|------|----------|------------|-------------|---------------|-------|
|RF-DETR|Nano |67.6 |48.4 |84.1 |57.1 |2.32 |
|RF-DETR|Small |72.1 |53.0 |85.9 |59.6 |3.52 |
|RF-DETR|Medium|73.6 |54.7 |86.6 |60.6 |4.52 |
|YOLO11|n |52.0 |37.4 |81.4 |55.3 |2.49 |
|YOLO11|s |59.7 |44.4 |82.3 |56.2 |3.16 |
|YOLO11|m |64.1 |48.6 |82.5 |56.5 |5.13 |
|YOLO11|l |65.3 |50.2 |x |x |6.65 |
|YOLO11|x |66.5 |51.2 |x |x |11.92 |
|LW-DETR|Tiny |60.7 |42.9 |x |x |1.91 |
|LW-DETR|Small |66.8 |48.0 |84.5 |58.0 |2.62 |
|LW-DETR|Medium|72.0 |52.6 |85.2 |59.4 |4.49 |
|D-FINE |Nano |60.2 |42.7 |83.6 |57.7 |2.12 |
|D-FINE |Small |67.6 |50.7 |84.5 |59.9 |3.55 |
|D-FINE |Medium|72.6 |55.1 |84.6 |60.2 |5.68 |
[See our benchmark notes in the RF-DETR documentation.](https://rfdetr.roboflow.com/learn/benchmarks/)
_We are actively working on RF-DETR Large and X-Large models using the same techniques we used to achieve the strong accuracy that RF-DETR Medium attains. This is why RF-DETR Large and X-Large is not yet reported on our pareto charts and why we haven't benchmarked other models at similar sizes. Check back in the next few weeks for the launch of new RF-DETR Large and X-Large models._
## Installation
To install RF-DETR, install the `rfdetr` package in a [**Python>=3.9**](https://www.python.org/) environment with `pip`:
```bash
pip install rfdetr
```
<details>
<summary>Install from source</summary>
<br>
By installing RF-DETR from source, you can explore the most recent features and enhancements that have not yet been officially released. Please note that these updates are still in development and may not be as stable as the latest published release.
```bash
pip install git+https://github.com/roboflow/rf-detr.git
```
</details>
## Inference
The easiest path to deployment is using Roboflow's [Inference](https://github.com/roboflow/inference) package.
The code below lets you run `rfdetr-base` on an image:
```python
import os
import supervision as sv
from inference import get_model
from PIL import Image
from io import BytesIO
import requests
url = "https://media.roboflow.com/dog.jpeg"
image = Image.open(BytesIO(requests.get(url).content))
model = get_model("rfdetr-base")
predictions = model.infer(image, confidence=0.5)[0]
detections = sv.Detections.from_inference(predictions)
labels = [prediction.class_name for prediction in predictions.predictions]
annotated_image = image.copy()
annotated_image = sv.BoxAnnotator(color=sv.ColorPalette.ROBOFLOW).annotate(annotated_image, detections)
annotated_image = sv.LabelAnnotator(color=sv.ColorPalette.ROBOFLOW).annotate(annotated_image, detections, labels)
```
## Predict
You can also use the .predict method to perform inference during local development. The `.predict()` method accepts various input formats, including file paths, PIL images, NumPy arrays, and torch tensors. Please ensure inputs use RGB channel order. For `torch.Tensor` inputs specifically, they must have a shape of `(3, H, W)` with values normalized to the `[0..1)` range. If you don't plan to modify the image or batch size dynamically at runtime, you can also use `.optimize_for_inference()` to get up to 2x end-to-end speedup, depending on platform.
```python
import io
import requests
import supervision as sv
from PIL import Image
from rfdetr import RFDETRBase
from rfdetr.util.coco_classes import COCO_CLASSES
model = RFDETRBase()
model = model.optimize_for_inference()
url = "https://media.roboflow.com/notebooks/examples/dog-2.jpeg"
image = Image.open(io.BytesIO(requests.get(url).content))
detections = model.predict(image, threshold=0.5)
labels = [
f"{COCO_CLASSES[class_id]} {confidence:.2f}"
for class_id, confidence
in zip(detections.class_id, detections.confidence)
]
annotated_image = image.copy()
annotated_image = sv.BoxAnnotator().annotate(annotated_image, detections)
annotated_image = sv.LabelAnnotator().annotate(annotated_image, detections, labels)
sv.plot_image(annotated_image)
```
### Train a Model
You can fine-tune an RF-DETR Nano, Small, Medium, and Base model with a custom dataset using the `rfdetr` Python package.
[Read our training tutorial to get started](https://rfdetr.roboflow.com/learn/train/)
## Documentation
Visit our [documentation website](https://rfdetr.roboflow.com) to learn more about how to use RF-DETR.
## License
Both the code and the weights pretrained on the COCO dataset are released under the [Apache 2.0 license](https://github.com/roboflow/r-flow/blob/main/LICENSE).
## Acknowledgements
Our work is built upon [LW-DETR](https://arxiv.org/pdf/2406.03459), [DINOv2](https://arxiv.org/pdf/2304.07193), and [Deformable DETR](https://arxiv.org/pdf/2010.04159). Thanks to their authors for their excellent work!
## Citation
If you find our work helpful for your research, please consider citing the following BibTeX entry.
```bibtex
@software{rf-detr,
author = {Robinson, Isaac and Robicheaux, Peter and Popov, Matvei},
license = {Apache-2.0},
title = {RF-DETR},
howpublished = {\url{https://github.com/roboflow/rf-detr}},
year = {2025},
note = {SOTA Real-Time Object Detection Model}
}
```
## Contribute
We welcome and appreciate all contributions! If you notice any issues or bugs, have questions, or would like to suggest new features, please [open an issue](https://github.com/roboflow/rf-detr/issues/new) or pull request. By sharing your ideas and improvements, you help make RF-DETR better for everyone.
<div align="center">
<a href="https://youtube.com/roboflow">
<img
src="https://media.roboflow.com/notebooks/template/icons/purple/youtube.png?ik-sdk-version=javascript-1.4.3&updatedAt=1672949634652"
width="3%"
/>
</a>
<img src="https://raw.githubusercontent.com/ultralytics/assets/main/social/logo-transparent.png" width="3%"/>
<a href="https://roboflow.com">
<img
src="https://media.roboflow.com/notebooks/template/icons/purple/roboflow-app.png?ik-sdk-version=javascript-1.4.3&updatedAt=1672949746649"
width="3%"
/>
</a>
<img src="https://raw.githubusercontent.com/ultralytics/assets/main/social/logo-transparent.png" width="3%"/>
<a href="https://www.linkedin.com/company/roboflow-ai/">
<img
src="https://media.roboflow.com/notebooks/template/icons/purple/linkedin.png?ik-sdk-version=javascript-1.4.3&updatedAt=1672949633691"
width="3%"
/>
</a>
<img src="https://raw.githubusercontent.com/ultralytics/assets/main/social/logo-transparent.png" width="3%"/>
<a href="https://docs.roboflow.com">
<img
src="https://media.roboflow.com/notebooks/template/icons/purple/knowledge.png?ik-sdk-version=javascript-1.4.3&updatedAt=1672949634511"
width="3%"
/>
</a>
<img src="https://raw.githubusercontent.com/ultralytics/assets/main/social/logo-transparent.png" width="3%"/>
<a href="https://discuss.roboflow.com">
<img
src="https://media.roboflow.com/notebooks/template/icons/purple/forum.png?ik-sdk-version=javascript-1.4.3&updatedAt=1672949633584"
width="3%"
/>
<img src="https://raw.githubusercontent.com/ultralytics/assets/main/social/logo-transparent.png" width="3%"/>
<a href="https://blog.roboflow.com">
<img
src="https://media.roboflow.com/notebooks/template/icons/purple/blog.png?ik-sdk-version=javascript-1.4.3&updatedAt=1672949633605"
width="3%"
/>
</a>
</a>
</div>
</div>
Raw data
{
"_id": null,
"home_page": null,
"name": "rfdetr",
"maintainer": null,
"docs_url": null,
"requires_python": ">=3.9",
"maintainer_email": null,
"keywords": "machine-learning, deep-learning, vision, ML, DL, AI, DETR, RF-DETR, Roboflow",
"author": null,
"author_email": "\"Roboflow, Inc\" <peter@roboflow.com>",
"download_url": "https://files.pythonhosted.org/packages/c6/ca/54d3839d947cb2b0c9b6ca2b68f2dd865e4f8d75ac7b8e78d7786a3c162e/rfdetr-1.2.1.tar.gz",
"platform": null,
"description": "# RF-DETR: SOTA Real-Time Object Detection Model\n\n[](https://badge.fury.io/py/rfdetr)\n[](https://pypistats.org/packages/rfdetr)\n[](https://badge.fury.io/py/rfdetr)\n[](https://github.com/roboflow/rfdetr/blob/main/LICENSE)\n\n[](https://huggingface.co/spaces/SkalskiP/RF-DETR)\n[](https://colab.research.google.com/github/roboflow-ai/notebooks/blob/main/notebooks/how-to-finetune-rf-detr-on-detection-dataset.ipynb)\n[](https://blog.roboflow.com/rf-detr)\n[](https://discord.gg/GbfgXGJ8Bk)\n\nRF-DETR is a real-time, transformer-based object detection model architecture developed by Roboflow and released under the Apache 2.0 license.\n\nRF-DETR is the first real-time model to exceed 60 AP on the [Microsoft COCO benchmark](https://cocodataset.org/#home) alongside competitive performance at base sizes. It also achieves state-of-the-art performance on [RF100-VL](https://github.com/roboflow/rf100-vl), an object detection benchmark that measures model domain adaptability to real world problems. RF-DETR is fastest and most accurate for its size when compared current real-time objection models.\n\nRF-DETR is small enough to run on the edge using [Inference](https://github.com/roboflow/inference), making it an ideal model for deployments that need both strong accuracy and real-time performance.\n\n[Read the documentation to get started training.](https://rfdetr.roboflow.com)\n\n## News\n\n- `2025/07/23`: We release three new checkpoints for RF-DETR: Nano, Small, and Medium.\n - RF-DETR Base is now deprecated. We recommend using RF-DETR Medium which offers subtantially better accuracy at comparable latency.\n- `2025/03/20`: We release RF-DETR real-time object detection model. **Code and checkpoint for RF-DETR-large and RF-DETR-base are available.**\n- `2025/04/03`: We release early stopping, gradient checkpointing, metrics saving, training resume, TensorBoard and W&B logging support.\n- `2025/05/16`: We release an 'optimize_for_inference' method which speeds up native PyTorch by up to 2x, depending on platform.\n\n## Results\n\nRF-DETR achieves state-of-the-art performance on both the Microsoft COCO and the RF100-VL benchmarks.\n\nThe table below shows the performance of RF-DETR medium, compared to comparable medium models:\n\n\n\n|family|size |coco_map50|coco_map50@95|rf100vl_map50|rv100vl_map50@95|latency|\n|------|------|----------|------------|-------------|---------------|-------|\n|RF-DETR|Nano |67.6 |48.4 |84.1 |57.1 |2.32 |\n|RF-DETR|Small |72.1 |53.0 |85.9 |59.6 |3.52 |\n|RF-DETR|Medium|73.6 |54.7 |86.6 |60.6 |4.52 |\n|YOLO11|n |52.0 |37.4 |81.4 |55.3 |2.49 |\n|YOLO11|s |59.7 |44.4 |82.3 |56.2 |3.16 |\n|YOLO11|m |64.1 |48.6 |82.5 |56.5 |5.13 |\n|YOLO11|l |65.3 |50.2 |x |x |6.65 |\n|YOLO11|x |66.5 |51.2 |x |x |11.92 |\n|LW-DETR|Tiny |60.7 |42.9 |x |x |1.91 |\n|LW-DETR|Small |66.8 |48.0 |84.5 |58.0 |2.62 |\n|LW-DETR|Medium|72.0 |52.6 |85.2 |59.4 |4.49 |\n|D-FINE |Nano |60.2 |42.7 |83.6 |57.7 |2.12 |\n|D-FINE |Small |67.6 |50.7 |84.5 |59.9 |3.55 |\n|D-FINE |Medium|72.6 |55.1 |84.6 |60.2 |5.68 |\n\n[See our benchmark notes in the RF-DETR documentation.](https://rfdetr.roboflow.com/learn/benchmarks/)\n\n_We are actively working on RF-DETR Large and X-Large models using the same techniques we used to achieve the strong accuracy that RF-DETR Medium attains. This is why RF-DETR Large and X-Large is not yet reported on our pareto charts and why we haven't benchmarked other models at similar sizes. Check back in the next few weeks for the launch of new RF-DETR Large and X-Large models._\n\n## Installation\n\nTo install RF-DETR, install the `rfdetr` package in a [**Python>=3.9**](https://www.python.org/) environment with `pip`:\n\n```bash\npip install rfdetr\n```\n\n<details>\n<summary>Install from source</summary>\n\n<br>\n\nBy installing RF-DETR from source, you can explore the most recent features and enhancements that have not yet been officially released. Please note that these updates are still in development and may not be as stable as the latest published release.\n\n```bash\npip install git+https://github.com/roboflow/rf-detr.git\n```\n\n</details>\n\n## Inference\n\nThe easiest path to deployment is using Roboflow's [Inference](https://github.com/roboflow/inference) package. \n\nThe code below lets you run `rfdetr-base` on an image:\n\n```python\nimport os\nimport supervision as sv\nfrom inference import get_model\nfrom PIL import Image\nfrom io import BytesIO\nimport requests\n\nurl = \"https://media.roboflow.com/dog.jpeg\"\nimage = Image.open(BytesIO(requests.get(url).content))\n\nmodel = get_model(\"rfdetr-base\")\n\npredictions = model.infer(image, confidence=0.5)[0]\n\ndetections = sv.Detections.from_inference(predictions)\n\nlabels = [prediction.class_name for prediction in predictions.predictions]\n\nannotated_image = image.copy()\nannotated_image = sv.BoxAnnotator(color=sv.ColorPalette.ROBOFLOW).annotate(annotated_image, detections)\nannotated_image = sv.LabelAnnotator(color=sv.ColorPalette.ROBOFLOW).annotate(annotated_image, detections, labels)\n```\n\n## Predict\n\nYou can also use the .predict method to perform inference during local development. The `.predict()` method accepts various input formats, including file paths, PIL images, NumPy arrays, and torch tensors. Please ensure inputs use RGB channel order. For `torch.Tensor` inputs specifically, they must have a shape of `(3, H, W)` with values normalized to the `[0..1)` range. If you don't plan to modify the image or batch size dynamically at runtime, you can also use `.optimize_for_inference()` to get up to 2x end-to-end speedup, depending on platform.\n\n```python\nimport io\nimport requests\nimport supervision as sv\nfrom PIL import Image\nfrom rfdetr import RFDETRBase\nfrom rfdetr.util.coco_classes import COCO_CLASSES\n\nmodel = RFDETRBase()\n\nmodel = model.optimize_for_inference()\n\nurl = \"https://media.roboflow.com/notebooks/examples/dog-2.jpeg\"\n\nimage = Image.open(io.BytesIO(requests.get(url).content))\ndetections = model.predict(image, threshold=0.5)\n\nlabels = [\n f\"{COCO_CLASSES[class_id]} {confidence:.2f}\"\n for class_id, confidence\n in zip(detections.class_id, detections.confidence)\n]\n\nannotated_image = image.copy()\nannotated_image = sv.BoxAnnotator().annotate(annotated_image, detections)\nannotated_image = sv.LabelAnnotator().annotate(annotated_image, detections, labels)\n\nsv.plot_image(annotated_image)\n```\n\n### Train a Model\n\nYou can fine-tune an RF-DETR Nano, Small, Medium, and Base model with a custom dataset using the `rfdetr` Python package.\n\n[Read our training tutorial to get started](https://rfdetr.roboflow.com/learn/train/)\n\n## Documentation\n\nVisit our [documentation website](https://rfdetr.roboflow.com) to learn more about how to use RF-DETR.\n\n## License\n\nBoth the code and the weights pretrained on the COCO dataset are released under the [Apache 2.0 license](https://github.com/roboflow/r-flow/blob/main/LICENSE).\n\n## Acknowledgements\n\nOur work is built upon [LW-DETR](https://arxiv.org/pdf/2406.03459), [DINOv2](https://arxiv.org/pdf/2304.07193), and [Deformable DETR](https://arxiv.org/pdf/2010.04159). Thanks to their authors for their excellent work!\n\n## Citation\n\nIf you find our work helpful for your research, please consider citing the following BibTeX entry.\n\n```bibtex\n@software{rf-detr,\n author = {Robinson, Isaac and Robicheaux, Peter and Popov, Matvei},\n license = {Apache-2.0},\n title = {RF-DETR},\n howpublished = {\\url{https://github.com/roboflow/rf-detr}},\n year = {2025},\n note = {SOTA Real-Time Object Detection Model}\n}\n```\n\n## Contribute\n\nWe welcome and appreciate all contributions! If you notice any issues or bugs, have questions, or would like to suggest new features, please [open an issue](https://github.com/roboflow/rf-detr/issues/new) or pull request. By sharing your ideas and improvements, you help make RF-DETR better for everyone.\n\n<div align=\"center\">\n <a href=\"https://youtube.com/roboflow\">\n <img\n src=\"https://media.roboflow.com/notebooks/template/icons/purple/youtube.png?ik-sdk-version=javascript-1.4.3&updatedAt=1672949634652\"\n width=\"3%\"\n />\n </a>\n <img src=\"https://raw.githubusercontent.com/ultralytics/assets/main/social/logo-transparent.png\" width=\"3%\"/>\n <a href=\"https://roboflow.com\">\n <img\n src=\"https://media.roboflow.com/notebooks/template/icons/purple/roboflow-app.png?ik-sdk-version=javascript-1.4.3&updatedAt=1672949746649\"\n width=\"3%\"\n />\n </a>\n <img src=\"https://raw.githubusercontent.com/ultralytics/assets/main/social/logo-transparent.png\" width=\"3%\"/>\n <a href=\"https://www.linkedin.com/company/roboflow-ai/\">\n <img\n src=\"https://media.roboflow.com/notebooks/template/icons/purple/linkedin.png?ik-sdk-version=javascript-1.4.3&updatedAt=1672949633691\"\n width=\"3%\"\n />\n </a>\n <img src=\"https://raw.githubusercontent.com/ultralytics/assets/main/social/logo-transparent.png\" width=\"3%\"/>\n <a href=\"https://docs.roboflow.com\">\n <img\n src=\"https://media.roboflow.com/notebooks/template/icons/purple/knowledge.png?ik-sdk-version=javascript-1.4.3&updatedAt=1672949634511\"\n width=\"3%\"\n />\n </a>\n <img src=\"https://raw.githubusercontent.com/ultralytics/assets/main/social/logo-transparent.png\" width=\"3%\"/>\n <a href=\"https://discuss.roboflow.com\">\n <img\n src=\"https://media.roboflow.com/notebooks/template/icons/purple/forum.png?ik-sdk-version=javascript-1.4.3&updatedAt=1672949633584\"\n width=\"3%\"\n />\n <img src=\"https://raw.githubusercontent.com/ultralytics/assets/main/social/logo-transparent.png\" width=\"3%\"/>\n <a href=\"https://blog.roboflow.com\">\n <img\n src=\"https://media.roboflow.com/notebooks/template/icons/purple/blog.png?ik-sdk-version=javascript-1.4.3&updatedAt=1672949633605\"\n width=\"3%\"\n />\n </a>\n </a>\n </div>\n</div>\n",
"bugtrack_url": null,
"license": "Apache License 2.0",
"summary": "RF-DETR",
"version": "1.2.1",
"project_urls": {
"Homepage": "https://github.com/roboflow/rf-detr"
},
"split_keywords": [
"machine-learning",
" deep-learning",
" vision",
" ml",
" dl",
" ai",
" detr",
" rf-detr",
" roboflow"
],
"urls": [
{
"comment_text": null,
"digests": {
"blake2b_256": "e7952c1db3eba8c6fda00a56e0027bbd8cb4128027e74d6a46e8ea0b1d309659",
"md5": "79daa75c142fc2c6bc02aca287ba106e",
"sha256": "ebbb201b25d0fd422d7cff02cdd75eddbe7221056f904b8e017821491cd460ca"
},
"downloads": -1,
"filename": "rfdetr-1.2.1-py3-none-any.whl",
"has_sig": false,
"md5_digest": "79daa75c142fc2c6bc02aca287ba106e",
"packagetype": "bdist_wheel",
"python_version": "py3",
"requires_python": ">=3.9",
"size": 131336,
"upload_time": "2025-07-24T11:54:02",
"upload_time_iso_8601": "2025-07-24T11:54:02.940621Z",
"url": "https://files.pythonhosted.org/packages/e7/95/2c1db3eba8c6fda00a56e0027bbd8cb4128027e74d6a46e8ea0b1d309659/rfdetr-1.2.1-py3-none-any.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": null,
"digests": {
"blake2b_256": "c6ca54d3839d947cb2b0c9b6ca2b68f2dd865e4f8d75ac7b8e78d7786a3c162e",
"md5": "a7a8620f99ff8dd0c644cf0e505431d2",
"sha256": "61f768a26e40d1ad19c6e609f2e699e9c691710edcaa52fd3ffd446956e52f0a"
},
"downloads": -1,
"filename": "rfdetr-1.2.1.tar.gz",
"has_sig": false,
"md5_digest": "a7a8620f99ff8dd0c644cf0e505431d2",
"packagetype": "sdist",
"python_version": "source",
"requires_python": ">=3.9",
"size": 113984,
"upload_time": "2025-07-24T11:54:04",
"upload_time_iso_8601": "2025-07-24T11:54:04.661357Z",
"url": "https://files.pythonhosted.org/packages/c6/ca/54d3839d947cb2b0c9b6ca2b68f2dd865e4f8d75ac7b8e78d7786a3c162e/rfdetr-1.2.1.tar.gz",
"yanked": false,
"yanked_reason": null
}
],
"upload_time": "2025-07-24 11:54:04",
"github": true,
"gitlab": false,
"bitbucket": false,
"codeberg": false,
"github_user": "roboflow",
"github_project": "rf-detr",
"travis_ci": false,
"coveralls": false,
"github_actions": true,
"lcname": "rfdetr"
}
JSON
