Ultralytics Yolo11 - CPU

Mix.install(
  [
    {:yolo, ">= 0.0.0"},
    {:nx, "~> 0.9"},
    {:exla, "~> 0.9"},
    {:image, "~> 0.54"},
    {:evision, "~> 0.2"},
    {:kino, "~> 0.16"},
    {:kino_yolo, github: "poeticoding/kino_yolo", branch: "main"},
    {:pythonx, "~> 0.4.2"},
    {:kino_pythonx, "~> 0.1.0"}
  ],
  config: [
    nx: [default_backend: EXLA.Backend]
  ]
)

Application.put_env(:ortex, Ortex.Native, features: [:cpu])

[project]
name = "ultralytics_to_onnx"
version = "0.0.0"
requires-python = "==3.13.*"
dependencies = [
  "ultralytics==8.3.155",
  "onnx==1.18.0",
  "onnxruntime==1.22.0",
  "onnx_pytorch==0.1.5",
  "requests"
]

Define model and paths

model_name = "yolo11m"

"yolo11m"

image_path = "benchmarks/images/traffic.jpg"
model_path = "#{model_name}.onnx"
classes_path = "models/coco_classes.json"
:ok

:ok

Ultralytics Yolo11 to ONNX

Pythonx.eval(
  """
  from ultralytics import YOLO
  IMAGE_SIZE = 640
  model_name = model_name.decode("utf-8")
  
  model = YOLO(model_name)
  model.export(format='onnx', imgsz=IMAGE_SIZE, opset=12)
  """,
  %{"model_name" => model_name}
)
:ok

Downloading https://github.com/ultralytics/assets/releases/download/v8.3.0/yolo11m.pt to 'yolo11m.pt'...
Ultralytics 8.3.155 🚀 Python-3.13.1 torch-2.7.1 CPU (Apple M3)
YOLO11m summary (fused): 125 layers, 20,091,712 parameters, 0 gradients, 68.0 GFLOPs

PyTorch: starting from 'yolo11m.pt' with input shape (1, 3, 640, 640) BCHW and output shape(s) (1, 84, 8400) (38.8 MB)
requirements: Ultralytics requirements ['onnx>=1.12.0,<1.18.0', 'onnxslim>=0.1.56'] not found, attempting AutoUpdate...
Requirement already satisfied: onnx<1.18.0,>=1.12.0 in ./venv/lib/python3.10/site-packages (1.17.0)
Requirement already satisfied: onnxslim>=0.1.56 in ./venv/lib/python3.10/site-packages (0.1.57)
Requirement already satisfied: protobuf>=3.20.2 in ./venv/lib/python3.10/site-packages (from onnx<1.18.0,>=1.12.0) (6.31.1)
Requirement already satisfied: numpy>=1.20 in ./venv/lib/python3.10/site-packages (from onnx<1.18.0,>=1.12.0) (2.2.6)
Requirement already satisfied: packaging in ./venv/lib/python3.10/site-packages (from onnxslim>=0.1.56) (25.0)
Requirement already satisfied: sympy in ./venv/lib/python3.10/site-packages (from onnxslim>=0.1.56) (1.14.0)
Requirement already satisfied: mpmath<1.4,>=1.1.0 in ./venv/lib/python3.10/site-packages (from sympy->onnxslim>=0.1.56) (1.3.0)

requirements: AutoUpdate success ✅ 0.8s
WARNING ⚠️ requirements: Restart runtime or rerun command for updates to take effect


ONNX: starting export with onnx 1.18.0 opset 12...
WARNING ⚠️ ONNX: simplifier failure: No module named 'onnxslim'
ONNX: export success ✅ 1.4s, saved as 'yolo11m.onnx' (76.9 MB)

Export complete (1.9s)
Results saved to /Users/alvise/code/yolo_elixir
Predict:         yolo predict task=detect model=yolo11m.onnx imgsz=640  
Validate:        yolo val task=detect model=yolo11m.onnx imgsz=640 data=/ultralytics/ultralytics/cfg/datasets/coco.yaml  
Visualize:       https://netron.app

:ok

Load image

mat = Evision.imread(image_path)

%Evision.Mat{
  channels: 3,
  dims: 2,
  type: {:u, 8},
  raw_type: 16,
  shape: {1080, 1920, 3},
  ref: #Reference<0.2594270970.3468558359.70281>
}

Load and run YOLO

model = YOLO.load(model_path: model_path, classes_path: classes_path, eps: [:cpu])

detected_objects = 
  model
  |> YOLO.detect(mat, iou_threshold: 0.45, prob_threshold: 0.25)
  |> YOLO.to_detected_objects(model.classes)

17:41:08.764 [info] Loaded model yolo11m.onnx with [:cpu] execution providers

17:41:08.765 [info] Initialized model yolo11m.onnx

[
  %{
    class: "person",
    prob: 0.3464750051498413,
    bbox: %{h: 95, w: 47, cx: 1675, cy: 391},
    class_idx: 0
  },
  %{
    class: "person",
    prob: 0.34952861070632935,
    bbox: %{h: 160, w: 57, cx: 1768, cy: 501},
    class_idx: 0
  },
  %{
    class: "person",
    prob: 0.36030319333076477,
    bbox: %{h: 157, w: 55, cx: 355, cy: 462},
    class_idx: 0
  },
  %{
    class: "person",
    prob: 0.5312483310699463,
    bbox: %{h: 110, w: 50, cx: 695, cy: 386},
    class_idx: 0
  },
  %{
    class: "person",
    prob: 0.5382186770439148,
    bbox: %{h: 155, w: 64, cx: 803, cy: 570},
    class_idx: 0
  },
  %{
    class: "person",
    prob: 0.5533820390701294,
    bbox: %{h: 165, w: 76, cx: 44, cy: 521},
    class_idx: 0
  },
  %{
    class: "person",
    prob: 0.5973995923995972,
    bbox: %{h: 141, w: 43, cx: 1617, cy: 421},
    class_idx: 0
  },
  %{
    class: "person",
    prob: 0.6149826049804688,
    bbox: %{h: 128, w: 42, cx: 1533, cy: 413},
    class_idx: 0
  },
  %{
    class: "person",
    prob: 0.6478251814842224,
    bbox: %{h: 120, w: 44, cx: 556, cy: 403},
    class_idx: 0
  },
  %{
    class: "person",
    prob: 0.7199457883834839,
    bbox: %{h: 157, w: 52, cx: 1685, cy: 491},
    class_idx: 0
  },
  %{
    class: "person",
    prob: 0.7302370667457581,
    bbox: %{h: 197, w: 94, cx: 733, cy: 687},
    class_idx: 0
  },
  %{
    class: "person",
    prob: 0.7496381402015686,
    bbox: %{h: 130, w: 69, cx: 701, cy: 567},
    class_idx: 0
  },
  %{
    class: "person",
    prob: 0.7714580297470093,
    bbox: %{h: 156, w: 64, cx: 302, cy: 558},
    class_idx: 0
  },
  %{
    class: "person",
    prob: 0.7862504720687866,
    bbox: %{h: 222, w: 90, cx: 469, cy: 846},
    class_idx: 0
  },
  %{
    class: "person",
    prob: 0.7874720096588135,
    bbox: %{h: 238, w: 81, cx: 41, cy: 801},
    class_idx: 0
  },
  %{
    class: "person",
    prob: 0.8138344883918762,
    bbox: %{h: 233, w: 101, cx: 679, cy: 848},
    class_idx: 0
  },
  %{
    class: "person",
    prob: 0.8157120943069458,
    bbox: %{h: 200, w: 81, cx: 605, cy: 768},
    class_idx: 0
  },
  %{
    class: "bicycle",
    prob: 0.4413682520389557,
    bbox: %{h: 83, w: 32, cx: 803, cy: 625},
    class_idx: 1
  },
  %{
    class: "bicycle",
    prob: 0.4464598298072815,
    bbox: %{h: 135, w: 56, cx: 599, cy: 849},
    class_idx: 1
  },
  %{
    class: "bicycle",
    prob: 0.4616367518901825,
    bbox: %{h: 182, w: 56, cx: 675, cy: 925},
    class_idx: 1
  },
  %{
    class: "bicycle",
    prob: 0.5315952897071838,
    bbox: %{h: 75, w: 42, cx: 694, cy: 416},
    class_idx: 1
  },
  %{
    class: "bicycle",
    prob: 0.6614882349967957,
    bbox: %{h: 165, w: 69, cx: 464, cy: 955},
    class_idx: 1
  },
  %{
    class: "bicycle",
    prob: 0.6782569885253906,
    bbox: %{h: 113, w: 63, cx: 725, cy: 737},
    class_idx: 1
  },
  %{class: "car", prob: 0.35281914472579956, bbox: %{h: 35, w: 57, cx: 1090, cy: 18}, class_idx: 2},
  %{
    class: "car",
    prob: 0.4985942840576172,
    bbox: %{h: 245, w: 224, cx: 1098, cy: 474},
    class_idx: 2
  },
  %{class: "car", prob: 0.5132410526275635, bbox: %{h: 64, w: 84, cx: 1034, cy: 167}, class_idx: 2},
  %{class: "car", prob: 0.5697806477546692, bbox: %{h: 71, w: 79, cx: 1172, cy: 149}, class_idx: 2},
  %{class: "car", prob: 0.6225624084472656, bbox: %{h: 49, w: 80, cx: 1193, cy: 187}, class_idx: 2},
  %{class: "car", prob: 0.652853786945343, bbox: %{h: 86, w: 97, cx: 1024, cy: 212}, class_idx: 2},
  %{class: "car", prob: 0.7560621500015259, bbox: %{h: 81, w: 105, cx: 1203, cy: 238}, class_idx: 2},
  %{
    class: "car",
    prob: 0.8117802143096924,
    bbox: %{h: 110, w: 129, cx: 1254, cy: 324},
    class_idx: 2
  },
  %{class: "car", prob: 0.8485523462295532, bbox: %{h: 88, w: 102, cx: 1038, cy: 268}, class_idx: 2},
  %{
    class: "car",
    prob: 0.8879761695861816,
    bbox: %{h: 121, w: 148, cx: 1301, cy: 402},
    class_idx: 2
  },
  %{
    class: "motorcycle",
    prob: 0.2533663511276245,
    bbox: %{h: 43, w: 28, cx: 866, cy: 228},
    class_idx: 3
  },
  %{
    class: "truck",
    prob: 0.580411970615387,
    bbox: %{h: 246, w: 223, cx: 1098, cy: 474},
    class_idx: 7
  },
  %{
    class: "traffic light",
    prob: 0.30687159299850464,
    bbox: %{h: 46, w: 44, cx: 1478, cy: 225},
    class_idx: 9
  },
  %{
    class: "traffic light",
    prob: 0.4421222507953644,
    bbox: %{h: 137, w: 75, cx: 864, cy: 113},
    class_idx: 9
  },
  %{
    class: "traffic light",
    prob: 0.6856074929237366,
    bbox: %{h: 104, w: 38, cx: 195, cy: 387},
    class_idx: 9
  },
  %{
    class: "traffic light",
    prob: 0.7127625346183777,
    bbox: %{h: 141, w: 77, cx: 1333, cy: 103},
    class_idx: 9
  },
  %{
    class: "backpack",
    prob: 0.3877813518047333,
    bbox: %{h: 67, w: 35, cx: 449, cy: 951},
    class_idx: 24
  },
  %{
    class: "backpack",
    prob: 0.6103652119636536,
    bbox: %{h: 58, w: 51, cx: 719, cy: 646},
    class_idx: 24
  },
  %{
    class: "backpack",
    prob: 0.6811151504516602,
    bbox: %{h: 76, w: 55, cx: 28, cy: 759},
    class_idx: 24
  },
  %{
    class: "handbag",
    prob: 0.3150494694709778,
    bbox: %{h: 45, w: 51, cx: 1768, cy: 488},
    class_idx: 26
  }
]

{:ok, image} = Image.from_evision(mat)
KinoYOLO.Draw.draw_detected_objects(image, detected_objects)