Add yolo26n_det in samples/linux/python

Francis235 · Francis235 · commit 59a118fb47ef · 2026-03-10T16:59:05.000+08:00
diff --git a/samples/linux/README.md b/samples/linux/README.md
@@ -159,6 +159,7 @@ The following table lists all available sample models with descriptions and exec
 | wideresnet50           | WideResNet50 image classification             | `python wideresnet50/wideresnet50.py`           |
 | xlsr                   | XLSR super-resolution                       | `python xlsr/xlsr.py`                           |
 | yolov8_det             | YOLOv8 object detection                       | `python yolov8_det/yolov8_det.py`               |
+| yolo26n_det            | YOLO26n object detection                      | `python yolov26n_det/yolo26n_det.py`            |
 
 > **Note:** Ensure you are in the `samples/linux/python` directory before running any example. Each sample will automatically download its required model on first run.
 
diff --git a/samples/linux/python/yolo26n_det/README.md b/samples/linux/python/yolo26n_det/README.md
@@ -0,0 +1,59 @@
+# yolo26n_det Sample Code
+
+## Introduction
+This is sample code for using AppBuilder to load yolo26n_det QNN model to HTP and execute inference to predicts bounding boxes and classes of objects in an image. The yolo26n_det.py file corresponds to the qai_appbuilder version. Additionly, we provide both the ultralytics version and onnxruntime version scripts. The ultralytics version script can be used to export an .onnx model.
+
+## How to get the QNN format model
+### 1. Export .onnx format model using ultralytics version script
+```bash
+python ultralytics_version.py
+```
+
+### 2. Set QNN related Environment Variables:
+```bash
+export QNN_SDK_ROOT=<path_to_your_qnn_sdk_root>
+export LD_LIBRARY_PATH=$QNN_SDK_ROOT/lib/aarch64-oe-linux-gcc11.2:$LD_LIBRARY_PATH
+export PYTHONPATH=${QNN_SDK_ROOT}/lib/python
+export PATH=${QNN_SDK_ROOT}/bin/x86_64-linux-clang/:$PATH
+export CPLUS_INCLUDE_PATH=/usr/include/c++/9:/usr/include/x86_64-linux-gnu/c++/9
+export LIBRARY_PATH=/usr/lib/gcc/x86_64-linux-gnu/9
+```
+
+### 3. Model Conversion
+```bash
+$QNN_SDK_ROOT/bin/x86_64-linux-clang/qnn-onnx-converter \
+    -i ./yolo26n.onnx \
+    --preserve_io layout \
+    --preserve_io datatype \
+    --output_path ./yolo26n.cpp
+```
+
+### 4. Model Lib Generation
+```bash
+$QNN_SDK_ROOT/bin/x86_64-linux-clang/qnn-model-lib-generator \
+    -c ./yolo26n.cpp \
+    -b ./yolo26n.bin \
+    -t aarch64-oe-linux-gcc11.2 \
+    -o ./
+```
+At this stage, a .so model library is generated. This library can be used on both the QNN CPU and HTP backends.
+
+### 5. Context Binary Generation
+```bash
+$QNN_SDK_ROOT/bin/x86_64-linux-clang/qnn-context-binary-generator \
+    --model ./x86_64-linux-clang/libyolo26n.so \
+    --soc_model 77 \
+    --backend ${QNN_SDK_ROOT}/lib/x86_64-linux-clang/libQnnHtp.so \
+    --binary_file cntx_yolo26n_soc77_fp16
+```
+At this stage, a context binary (.bin) is generated, which can only be used on the HTP backend.
+
+For more information, please refer to the QAIRT SDK documentation.
+
+## Notes
+- When running the `yolo26n_det/yolo26n_det.py` script, make sure to replace the `model_path` with your own model path generated in the previous step.
+
+## References
+[1] https://docs.qualcomm.com/nav/home/general_tools.html?product=1601111740009302
+
+[2] https://docs.ultralytics.com/models/yolo26/
diff --git a/samples/linux/python/yolo26n_det/input.jpg b/samples/linux/python/yolo26n_det/input.jpg
diff --git a/samples/linux/python/yolo26n_det/onnxruntime_version.py b/samples/linux/python/yolo26n_det/onnxruntime_version.py
@@ -0,0 +1,184 @@
+import argparse
+
+import cv2
+import numpy as np
+import onnxruntime as ort
+
+
+IMAGE_SIZE = 640  # model expects [1,3,640,640]
+
+# define class type (COCO 80)
+class_map = {
+    0: "person",
+    1: "bicycle",
+    2: "car",
+    3: "motorcycle",
+    4: "airplane",
+    5: "bus",
+    6: "train",
+    7: "truck",
+    8: "boat",
+    9: "traffic light",
+    10: "fire hydrant",
+    11: "stop sign",
+    12: "parking meter",
+    13: "bench",
+    14: "bird",
+    15: "cat",
+    16: "dog",
+    17: "horse",
+    18: "sheep",
+    19: "cow",
+    20: "elephant",
+    21: "bear",
+    22: "zebra",
+    23: "giraffe",
+    24: "backpack",
+    25: "umbrella",
+    26: "handbag",
+    27: "tie",
+    28: "suitcase",
+    29: "frisbee",
+    30: "skis",
+    31: "snowboard",
+    32: "sports ball",
+    33: "kite",
+    34: "baseball bat",
+    35: "baseball glove",
+    36: "skateboard",
+    37: "surfboard",
+    38: "tennis racket",
+    39: "bottle",
+    40: "wine glass",
+    41: "cup",
+    42: "fork",
+    43: "knife",
+    44: "spoon",
+    45: "bowl",
+    46: "banana",
+    47: "apple",
+    48: "sandwich",
+    49: "orange",
+    50: "broccoli",
+    51: "carrot",
+    52: "hot dog",
+    53: "pizza",
+    54: "donut",
+    55: "cake",
+    56: "chair",
+    57: "couch",
+    58: "potted plant",
+    59: "bed",
+    60: "dining table",
+    61: "toilet",
+    62: "tv",
+    63: "laptop",
+    64: "mouse",
+    65: "remote",
+    66: "keyboard",
+    67: "cell phone",
+    68: "microwave",
+    69: "oven",
+    70: "toaster",
+    71: "sink",
+    72: "refrigerator",
+    73: "book",
+    74: "clock",
+    75: "vase",
+    76: "scissors",
+    77: "teddy bear",
+    78: "hair drier",
+    79: "toothbrush"
+}
+
+
+def _select_providers() -> list[str]:
+    available = ort.get_available_providers()
+    preferred = ["CUDAExecutionProvider", "CPUExecutionProvider"]
+    return [p for p in preferred if p in available] or available
+
+
+def _load_image(image_path: str) -> tuple[np.ndarray, np.ndarray]:
+    img0 = cv2.imread(image_path)
+    if img0 is None:
+        raise FileNotFoundError(f"Failed to read image: {image_path}")
+
+    img = cv2.resize(img0, (IMAGE_SIZE, IMAGE_SIZE), interpolation=cv2.INTER_LINEAR)
+    img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
+    img = img.astype(np.float32) / 255.0
+    img = np.transpose(img, (2, 0, 1))[None]  # NCHW
+    return img0, img
+
+
+def _draw_xyxy(
+    frame_bgr: np.ndarray,
+    xyxy_640: np.ndarray,
+    score: float,
+    class_id: int,
+    color: tuple[int, int, int] = (0, 255, 0),
+    thickness: int = 2,
+) -> None:
+    h, w = frame_bgr.shape[:2]
+    scale_x = w / IMAGE_SIZE
+    scale_y = h / IMAGE_SIZE
+
+    x1, y1, x2, y2 = xyxy_640.tolist()
+    x1 = int(max(0.0, min(IMAGE_SIZE, x1)) * scale_x)
+    y1 = int(max(0.0, min(IMAGE_SIZE, y1)) * scale_y)
+    x2 = int(max(0.0, min(IMAGE_SIZE, x2)) * scale_x)
+    y2 = int(max(0.0, min(IMAGE_SIZE, y2)) * scale_y)
+
+    cv2.rectangle(frame_bgr, (x1, y1), (x2, y2), color, thickness)
+    class_name = class_map.get(class_id, "Unknown")
+    label = f"{score:.2f} {class_name}"
+    cv2.putText(
+        frame_bgr,
+        label,
+        (x1, max(0, y1 - 8)),
+        cv2.FONT_HERSHEY_SIMPLEX,
+        0.6,
+        color,
+        2,
+        cv2.LINE_AA,
+    )
+
+
+def main() -> int:
+    parser = argparse.ArgumentParser(description="YOLO26 ONNXRuntime inference (no NMS).")
+    parser.add_argument("--model", default="yolo26n.onnx", help="Path to YOLO26 ONNX model")
+    parser.add_argument("--image", default="input.jpg", help="Path to input image")
+    parser.add_argument("--output", default="output.jpg", help="Path to save visualization")
+    parser.add_argument("--conf", type=float, default=0.25, help="Confidence threshold")
+    args = parser.parse_args()
+
+    providers = _select_providers()
+    try:
+        sess = ort.InferenceSession(args.model, providers=providers)
+    except Exception:
+        sess = ort.InferenceSession(args.model, providers=["CPUExecutionProvider"])
+        providers = ["CPUExecutionProvider"]
+
+    input_name = sess.get_inputs()[0].name
+
+    img0, inp = _load_image(args.image)
+    (out,) = sess.run(None, {input_name: inp})
+
+    det = np.asarray(out, dtype=np.float32)[0]  # [300,6]
+    scores = det[:, 4]
+    keep = scores >= float(args.conf)
+    det = det[keep]
+    det = det[np.argsort(-det[:, 4])]
+
+    for x1, y1, x2, y2, score, cls in det:
+        _draw_xyxy(img0, np.array([x1, y1, x2, y2], dtype=np.float32), float(score), int(cls))
+
+    if not cv2.imwrite(args.output, img0):
+        raise RuntimeError(f"Failed to write image: {args.output}")
+
+    print(f"onnxruntime={ort.__version__} providers={providers}")
+    print(f"detections={len(det)} saved={args.output}")
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())
diff --git a/samples/linux/python/yolo26n_det/ultralytics_version.py b/samples/linux/python/yolo26n_det/ultralytics_version.py
@@ -0,0 +1,13 @@
+from ultralytics import YOLO
+
+# Load the YOLO26 model
+model = YOLO("yolo26n.pt or your .pt model path")
+
+# Export the model to ONNX format
+model.export(format="onnx")  # creates 'yolo26n.onnx'
+
+# Load the exported ONNX model
+onnx_model = YOLO("your .onnx model path")
+
+# Run inference
+results = onnx_model("your input image")
diff --git a/samples/linux/python/yolo26n_det/yolo26n_det.py b/samples/linux/python/yolo26n_det/yolo26n_det.py
