Roboflow + Hugging Face
Import Hugging Face Hub datasets into Roboflow, and load RF-DETR models, including ones you train on Roboflow, with the Hugging Face Transformers library.
How to connect Hugging Face to Roboflow
4 stepsClone a dataset from the Hub
Install git and git-lfs, then run git clone https://huggingface.co/datasets/<owner>/<dataset>. The dataset needs annotation files (XML, JSON or CSV) for labels to come across.
Upload it to a Roboflow project
Create a project with the matching task type, such as object detection, and upload the images and annotation files.
Train a model
Generate a version and train RF-DETR or another architecture in Roboflow.
Load RF-DETR in Transformers
Use AutoModelForObjectDetection or the object-detection pipeline with a checkpoint such as Roboflow/rf-detr-medium. Transformers also accepts RF-DETR weights trained on the Roboflow platform.
What teams build with Hugging Face
Start from a public Hub dataset
Clone a labeled dataset from Hugging Face, fix or extend its labels in Roboflow Annotate, and train on it.
Run a Roboflow-trained model in Transformers
Train RF-DETR on Roboflow and load the weights with the RfDetr classes in your own Transformers code.
Take a fine-tuned VLM elsewhere
Download a PaliGemma, Florence-2 or Qwen model fine-tuned on Roboflow as a standard Hugging Face directory.
Questions
There is no one-click push. Export the dataset version in the format you need and upload it to the Hub yourself.
Under the Roboflow organization on Hugging Face, for example Roboflow/rf-detr-medium and the rf-detr-seg models for segmentation. They are Apache 2.0 licensed.
Ones that ship annotation files in a format Roboflow reads, such as COCO JSON, Pascal VOC XML or CSV. Datasets stored only as Parquet tables need converting first.
Yes. Some models, such as Depth Anything V2, pull their weights from Hugging Face, so you set a HUGGING_FACE_HUB_TOKEN before running them with the inference package.
Why are companies choosing Roboflow?
The end-to-end platform for vision AI: create datasets, train models and deploy to production.
Flexible deployment modes
Move from a scalable cloud API to edge hardware like NVIDIA Jetson without managing complex dependencies.
Integrated workflow builder
Combine multiple AI models, custom business logic and third-party integrations into one production-ready pipeline in a low-code editor.
Model monitoring and alerts
Dashboards track confidence, inference time and request volume for your deployed models, and email alerts tell your team when confidence suddenly drops.
