Your First Detection

The vision_rs::detect module is the entry point for running inference. It wraps model-specific configuration and forward passes behind a single ObjectDetector/DetectorConfig pair, so callers don’t need to touch the model internals for simple inference.

use vision_rs::detect::{DetectorConfig, ObjectDetector, Yolo26DetectorConfig};
use vision_rs::models::yolo::yolo26::Yolo26Variant;

let config = DetectorConfig::Yolo26(Yolo26DetectorConfig::new(
    Yolo26Variant::N,
    "weights/yolo26n.bin",
    vec!["car".into(), "truck".into(), "person".into()],
));

let detector = ObjectDetector::new(config)?;
let image_bytes = std::fs::read("frame.jpg")?;
let detections = detector.detect(&image_bytes).await?;

for d in &detections {
    println!("{} {:.2} {:?}", d.class, d.confidence, d.bbox);
}

Walking through it

1. Pick a variant. Yolo26Variant has five sizes — N (nano) through XL — trading accuracy for speed/memory. See The YOLO26 Model for how they differ.
2. Build a config. Yolo26DetectorConfig::new takes the variant, a weights file path, and the class label list (must match the model’s training classes, in class-index order). It fills in sensible defaults: conf_threshold: 0.25, nms_iou_threshold: 0.45, img_size: 640. Adjust these fields directly on the returned config if your use case needs different thresholds.
3. Construct the detector. ObjectDetector::new takes ownership of the config and loads the model.
4. Run inference. detect() takes raw JPEG or PNG bytes and returns every Detection that clears conf_threshold, after non-maximum suppression. Each Detection has a bbox: [cx, cy, w, h] (normalised to [0, 1]), a resolved class label, and a confidence score.

Getting a model

The yolo26 example binary (examples/yolo26.rs) has a download subcommand for fetching pretrained weights and a dataset config, and a bench subcommand for a throughput/latency smoke test:

source .env
cargo build --release --example yolo26 --features cuda
./target/release/examples/yolo26 download --dataset assets/datasets/coco128.toml
./target/release/examples/yolo26 bench \
  --model ultralytics/yolo26n \
  --dataset assets/datasets/coco128.toml \
  --skip-map --warmup 10 --runs 100

See Benchmarking & Profiling for the full set of yolo26 subcommands and profiling workflows.