D-FINE Tutorial

This tutorial will guide you through using trtutils with D-FINE models. We will cover:

1. Downloading ONNX weights from D-FINE

2. Building a TensorRT engine

3. Running inference with the engine

4. Advanced features and optimizations

Downloading ONNX Weights

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

# Download and convert D-FINE models to ONNX
# Available models: dfine_n, dfine_s, dfine_m, dfine_l, dfine_x
$ python3 -m trtutils download --model dfine_n --output dfine_n.onnx --imgsz 640 --opset 17

# For other D-FINE variants
$ python3 -m trtutils download --model dfine_s --output dfine_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="dfine_n.onnx",
    output="dfine_n.engine",
    fp16=True,
    shapes=[("images", (1, 3, 640, 640)), ("orig_image_size", (1, 2))],
)

# For Jetson devices with DLA support
build_engine(
    onnx="dfine_n.onnx",
    output="dfine_n.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))],
)