nanodet-rs

Rust implementation of the nanodet decoder.

Supports multiple neural network backends via AsFeatureMatrix trait. Implementations are provided for the following backends:

• ncnn-rs - Works on most CPUs and has experimental Vulkan GPU support.
• openvino-rs - A very fast framework that utilises Intel integrated CPU graphics.

Usage

Can be used as library by including in Cargo.toml:

nanodet-rs = { git = "https://github.com/lit-robotics/nanodet-rs" }

For rust-flavored usage see examples/ncnn_image.rs, or if you like opencv check examples/ncnn_opencv.rs

Cargo Features

• ncnn - ncnn-rs support.
• openvino - openvino-rs support.
• image - image based utility functions.
• opencv - opencv based utility functions.

Running examples

ncnn

Run pretrained COCO model on an image with rust image pipeline:

cargo run --example ncnn_image --release --features ncnn,image -- data/coco_test.jpg

Run pretrained COCO model on an image with opencv image pipeline:

cargo run --example ncnn_opencv --release --features ncnn,opencv -- data/coco_test.jpg

openvino

Run pretrained COCO model on an image with opencv image pipeline:

# Initialize openvino environment (check openvino docs)
source /opt/intel/openvino_2022/setupvars.sh
# Run example
cargo run --example openvino_opencv --release --features openvino,opencv -- data/coco_test.jpg