RF-DETR Tutorial¶

This tutorial will guide you through using trtutils with RF-DETR models. We will cover:

Downloading ONNX weights from RF-DETR
Building a TensorRT engine
Running inference with the engine
Advanced features and optimizations

Downloading ONNX Weights¶

RF-DETR models can be automatically downloaded and converted to ONNX format using the trtutils CLI:

# Download and convert RF-DETR models to ONNX
# Available models: rfdetr_n, rfdetr_s, rfdetr_m
$ python3 -m trtutils download --model rfdetr_n --output rfdetr_n.onnx --imgsz 640 --opset 17

# For other RF-DETR variants
$ python3 -m trtutils download --model rfdetr_s --output rfdetr_s.onnx --imgsz 640 --opset 17

Building TensorRT Engine¶

Once you have the ONNX weights, you can build a TensorRT engine using trtutils:

# Note we can build directly from the ONNX weights
# Note: Need to specify the input shapes (namely the batch dimension is
# left dynamic)
python3 -m trtutils build \
    --onnx $ONNX_PATH \
    --output $OUTPUT_PATH \
    --fp16 \
    --shape images:1,3,640,640 \
    --shape orig_image_size:1,2

Alternatively, if you want to export the engine using the Python API:

from trtutils.builder import build_engine

# Build the engine with FP16 precision
build_engine(
    onnx="rfdetr_n.onnx",
    output="rfdetr_n.engine",
    fp16=True,
    shapes=[("images", (1, 3, 640, 640)), ("orig_image_size", (1, 2))],
)

# For Jetson devices with DLA support
build_engine(
    onnx="rfdetr_n.onnx",
    output="rfdetr_n_dla.engine",
    int8=True,  # Orin series optimize for int8
    fp16=True,  # Can use fp16 on Xavier series
    dla_core=0,  # Use DLA core 0
    shapes=[("images", (1, 3, 640, 640)), ("orig_image_size", (1, 2))],
)

Running Inference¶

The RFDETR class provides a high-level interface for running RF-DETR inference:

import cv2
from trtutils.models import RFDETR

# Load the RF-DETR model
rfdetr = RFDETR("rfdetr_n.engine")

# Read and process an image
img = cv2.imread("example.jpg")
detections = rfdetr.end2end(img)

# Print results
for bbox, confidence, class_id in detections:
    print(f"Class: {class_id}, Confidence: {confidence}")
    print(f"Bounding Box: {bbox}")