easy-rfdetr 0.1.0

YOLO-like API for RF-DETR real-time object detection

easy-rfdetr

Train a Model

from easy_rfdetr import RFDETR

model = RFDETR("medium")
model.train(data="my_dataset/", epochs=50)

Run Inference

model("image.jpg").show()

Install

pip install easy-rfdetr

Drop-in replacement for YOLO() from ultralytics. Same API, transformer accuracy.

---

Training

Drop your dataset in a folder. We auto-detect COCO or YOLO format:

dataset/
├── train/images/ + labels/
├── valid/images/ + labels/
└── test/images/ + labels/

Train:

model.train(data="dataset/", epochs=50, batch=8)

Resume:

model.train(data="dataset/", resume=True)

---

Inference

model = RFDETR("medium")  # nano, small, medium, large

# From file
model("photo.jpg").show()

# From URL
model("https://example.com/img.jpg")

# Batch
model(["a.jpg", "b.jpg"])

# Confidence threshold
model("img.jpg", threshold=0.8)

# Get boxes
r = model("img.jpg")
print(r.boxes)     # xyxy
print(r.scores)    # confidence
print(r.labels)    # ["person", "car", ...]

# Save
r.save("output.jpg")

---

Web UI

model.ui()

---

CLI

rfdetr predict source=image.jpg
rfdetr train data=dataset/ epochs=50

---

Requirements

• Python >= 3.10
• PyTorch >= 2.0.0
• CUDA or Apple Silicon (optional)

---

License

Apache 2.0 - See LICENSE

---

Built on RF-DETR by Roboflow.