Draw detected boxes on the image and recognized texts on a white panel. Args: image (np.ndarray): Original image. boxes (list): L

Log and draw detection results from `(boxes, scores, cls_ids)` arrays.

* @brief Draw detection results on image * @param img Image to draw on * @param detections Detection results * @param x_scale X scale factor for co

Overlay a small class legend on the top-right corner of canvas.

Draw pose estimation results.

* @brief Draw pose detection results */

Draw YOLOWorld detection results on the original image. Args: image: Original BGR image. boxes: Detection boxes in original

Draw rotated bounding boxes (OBB).

Draw CLIP similarity scores on the input image. Args: image: Original BGR image. texts: Candidate text prompts. scor

Execute the ONNX text encoder for a list of prompts. Args: texts: Prompt list to encode. Returns: T

Export YOLO26 model to BPU-friendly ONNX.

Export YOLO26 Classification model.

Export YOLO26 OBB model. Args: model_path (str): Path to input .pt file. output_name (str): Path to save output .onnx file.

Export YOLO26 Pose model.

Export YOLO26 Segmentation model.

Execute one forward pass on BPU. Args: inputs: Prepared input tensors returned by `pre_process()`. Returns:

Execute one forward pass on BPU. Args: inputs: Prepared input tensors returned by `pre_process()`. Returns:

Run BPU inference and return the raw logits tensor. Args: input_tensors (Dict[str, Dict[str, np.ndarray]]): Runtime inpu

Execute inference on BPU using hbm_runtime. Args: inputs (Dict[str, Dict[str, np.ndarray]]): Prepared input tensors.

Execute one forward pass on BPU. Args: inputs: Prepared input tensors returned by `pre_process()`. Returns:

Execute one forward pass on BPU. Args: inputs: Prepared input tensors returned by `pre_process()`. Returns:

Execute one forward pass on BPU. Args: inputs: Prepared input tensors returned by `pre_process()`. Returns:

Execute the image encoder on BPU. Args: inputs: Prepared image encoder inputs. Returns: Image featu

Execute one forward pass on BPU. Args: inputs: Prepared input tensors returned by `pre_process()`. Returns:

Execute one forward pass on BPU. Args: inputs: Prepared input tensors returned by `pre_process()`. Returns:

Execute inference on BPU using hbm_runtime. Args: input_tensor: Prepared input tensors. Returns: Raw output

Execute detection model inference on BPU. Args: input_tensor (Dict[str, Dict[str, np.ndarray]]): Prepared input tensors.

Run BPU inference. Args: nv12: Preprocessed NV12 array. Returns: Raw output array of shape (1, H, W, 1) int3

Execute one forward pass on BPU. Args: inputs: Prepared input tensors returned by `pre_process()`. Returns:

Execute inference on BPU using `hbm_runtime`. Args: inputs (Dict[str, Dict[str, np.ndarray]]): Runtime input dictionary

Execute one forward pass on BPU. Args: inputs: Prepared input tensors returned by `pre_process()`. Returns:

Execute one forward pass on BPU. Args: inputs: Prepared input tensors returned by `pre_process()`. Returns:

Execute one forward pass on BPU. Args: inputs: Prepared input tensors returned by `pre_process()`. Returns:

Execute one BPU inference pass. Args: inputs: Prepared image and text embedding tensors. Returns: R

Execute one forward pass on BPU. Args: inputs: Prepared input tensors returned by `pre_process()`. Returns:

Execute one forward pass on BPU. Args: inputs: Prepared input tensors returned by `pre_process()`. Returns:

Execute one forward pass on BPU. Args: inputs: Prepared input tensors returned by `pre_process()`. Returns:

Execute one forward pass on BPU. Args: inputs: Prepared input tensors returned by `pre_process()`. Returns:

Execute one forward pass on BPU. Args: inputs: Prepared input tensors returned by `pre_process()`. Returns:

Executes inference on BPU using hbm_runtime. Args: inputs (Dict): Prepared input tensors from pre_process(). Re

Execute one forward pass on BPU. Args: inputs: Prepared input tensors returned by `pre_process()`. Returns:

Execute one forward pass on BPU. Args: inputs: Prepared input tensors returned by `pre_process()`. Returns:

Run BPU inference for one input image. Args: input_tensor: Input tensor dictionary returned by `pre_process`. Returns:

Run BPU inference for one input image.

Run BPU inference for one input image.

Execute one forward pass on BPU. Args: inputs: Prepared input tensors returned by `pre_process()`. Returns:

Execute one forward pass on BPU. Args: inputs: Prepared input tensors returned by `pre_process()`. Returns:

Execute inference on BPU using hbm_runtime. Args: input_tensor: Prepared input tensors. Returns: Raw output

* @brief Get top K classification results */

从 CSV 文件加载 ground truth，返回 {相对路径字符串: 标签} 的字典。 第一列保持原样（不做任何 basename 提取），第二列为标签（0‑based）。 自动跳过标题行（如果第一行包含 'image' 和 'category'）。

Run the complete EfficientViT classification pipeline on a single image. The entry follows the standardized sample pattern used in this repo

Run the complete EdgeNeXt classification pipeline on a single image. The entry follows the standardized sample pattern used in this reposito

Run the complete LPRNet recognition pipeline on one binary input tensor. The entry follows the standardized sample pattern used in this repo

Run the complete YOLOv5 detection pipeline on a single image. This function parses command-line arguments, initializes the runtime wrapp

Run the complete ResNet classification pipeline on a single image. The entry follows the standardized sample pattern used in this repository

Run the complete RepVGG classification pipeline on a single image. The entry follows the standardized sample pattern used in this repository

Run the complete MobileNetV3 classification pipeline on a single image. The entry follows the standardized sample pattern used in this repos

Run CLIP image-text matching on one image. The entry follows the standard sample pattern: parse arguments, construct the runtime wrapper

Run the complete MobileNetV2 classification pipeline on a single image. The entry follows the standardized sample pattern used in this repos

Run the complete VargConvNet classification pipeline on a single image. The entry follows the standardized sample pattern used in this repos

Run the complete YOLOE segmentation pipeline on a single image.

Run the complete PaddleOCR pipeline on a single image. The entry follows the standardized sample pattern used in this repository: 1. Parse de

Run PP-LiteSeg-STDC1 inference on a single image.

Run the complete MobileNetV1 classification pipeline on a single image. The entry follows the standardized sample pattern used in this repos

Run the complete FCOS detection pipeline on a single image. The entry follows the standardized sample pattern used in this repository: 1

Run the complete RepGhost classification pipeline on a single image. The entry follows the standardized sample pattern used in this reposito

Run the complete EfficientFormer classification pipeline on a single image. The entry follows the standardized sample pattern used in this r

Run the complete MobileOne classification pipeline on a single image. The entry follows the standardized sample pattern used in this reposit

Run YOLOWorld open-vocabulary detection on one image. The entry follows the standard sample pattern: parse arguments, construct the wrap

Run the complete FasterNet classification pipeline on a single image. The entry follows the standardized sample pattern used in this reposit

Run the complete HGNetV2 classification pipeline on a single image. The entry follows the standardized sample pattern used in this repositor

Run the complete EfficientNet classification pipeline on a single image. The entry follows the standardized sample pattern used in this repo

Run the complete FastViT classification pipeline on a single image. The entry follows the standardized sample pattern used in this repositor

Run the complete ResNeXt classification pipeline on a single image. The entry follows the standardized sample pattern used in this repositor

Run ConvNeXt inference on a single image. This function orchestrates the complete inference process: - Argument parsing - Model initi

Run the complete EfficientFormerV2 classification pipeline on one image. The entry follows the standardized sample pattern used in this repo

Main execution block for model conversion.

Main entry point for detection model export.

Main entry point for OBB export.

Main entry point for segmentation model export.

Main entry point for classification model export.