@brief Run OBB detection on an image (optimized with buffer reuse) @param image Input image (BGR format) @param confThreshold Confidence threshold @pa

* @brief Print usage information */

@brief Draw OBB detections on an image @param image Image to draw on @param results Vector of OBB detection results @param thickness Line thickness

@brief Get scale and padding info from letterbox operation @param originalSize Original image size @param letterboxSize Letterboxed image size @param[

* @brief Print performance metrics * @param taskName Name of the task * @param durationMs Duration in milliseconds * @param fps Frames per second (

* @brief Save image with timestamp * @param image Image to save * @param inputPath Original input path * @param outputDir Output directory * @retu

@brief Run classification on an image @param image Input image (BGR format) @return Classification result

Get the first element from the queue.

@brief Draw classification result on an image

@brief Ensure blob has required capacity

@brief Load class names from a file (one class name per line) @param path Path to the class names file @return Vector of class names

* @brief Save video writer output path with timestamp * @param inputPath Original input path * @param outputDir Output directory * @return Output p

* @brief Check if file extension is supported image format */

@brief Run segmentation on an image (optimized with buffer reuse) @param image Input image (BGR format) @param confThreshold Confidence threshold @par

@brief Compute area of the bounding box

@brief Draw pose estimations on an image @param image Image to draw on @param results Vector of pose results @param kptRadius Keypoint circle radius @

@brief Draw segmentations with boxes and labels on an image @param image Image to draw on @param results Vector of segmentation results @param maskAlp

Add an element to the queue.

@brief Generate consistent random colors for each class @param classNames Vector of class names @param seed Random seed for reproducibility @return Ve

@brief Fast letterbox with direct blob output (avoids intermediate copies) @param image Input BGR image @param blob Output CHW float blob (pre-allocat

@brief Perform class-aware NMS by offsetting boxes by class ID @param boxes Vector of bounding boxes @param scores Vector of confidence scores @param

@brief Draw a single bounding box with label on an image @param image Image to draw on @param box Bounding box @param label Text label @param color Bo

@brief Draw detections with semi-transparent mask fill

* @brief Normalize angle to [-PI, PI] range */

============================================================================ Command-line Parsing ====================================================

@brief Perform Non-Maximum Suppression on bounding boxes @param boxes Vector of bounding boxes @param scores Vector of confidence scores @param scoreT

============================================================================ Benchmark Functions =====================================================

@brief Draw pose keypoints and skeleton on an image @param image Image to draw on @param keypoints Vector of keypoints @param skeleton Skeleton connec

@brief Draw a segmentation mask on an image @param image Image to draw on @param mask Binary mask (CV_8UC1) @param color Mask color @param alpha Mask

* @brief Generate output filename with timestamp * @param inputPath Original input file path * @param outputDir Output directory (e.g., "outputs/det

@brief Compute the product of elements in a vector @param shape Vector of dimensions @return Product of all elements

@brief Perform NMS on float-precision bounding boxes (for letterbox space) @param boxes Vector of cv::Rect2f boxes @param scores Vector of confidence

@brief Perform class-aware NMS on float-precision boxes

@brief Perform NMS on oriented bounding boxes using rotated IoU @param boxes Vector of oriented bounding boxes @param scores Vector of confidence scor

@brief Perform class-aware NMS on oriented bounding boxes @param boxes Vector of oriented bounding boxes @param scores Vector of confidence scores @pa

@brief Compute IoU between two oriented bounding boxes using OpenCV @param box1 First oriented bounding box @param box2 Second oriented bounding box @

@brief Detect YOLO version from detection model output tensor shape @param outputShape The shape of the first output tensor [batch, dim1, dim2, ...] @

@brief Draw a bounding box with semi-transparent mask fill @param image Image to draw on @param box Bounding box @param label Text label @param color

@brief Draw classification result on an image @param image Image to draw on @param result Classification result @param position Position for the text

@brief Draw an oriented bounding box on an image @param image Image to draw on @param obb Oriented bounding box @param label Text label @param color B

* @brief Get current timestamp string in format: YYYYMMDD_HHMMSS * @return Timestamp string */

* @brief Check if two OBBs match, considering angle ambiguity * * Two oriented bounding boxes can represent the same physical box with: * 1. Same

Preprocess image using OpenCV to match C++ implementation exactly. - Load BGR - Convert to RGB - Resize shortest side to target_size

Quantizes an ONNX model and saves the quantized version. Args: onnx_model_path: Path to the original ONNX model file. quanti