ComfyOnline
ONNXDetectorProvider

Introduction:

This document provides an overview of the ONNXDetectorProvider node within the ComfyUI-Impact-Pack, a custom node pack designed to enhance image processing workflows in ComfyUI. The Impact Pack offers a suite of tools focusing on detection, detailing, upscaling, and more. This document specifically focuses on the ONNXDetectorProvider.

ComfyUI-Impact-Pack Introduction:

The ComfyUI-Impact-Pack is a collection of custom nodes for ComfyUI, aiming to simplify and enhance image manipulation workflows. It provides tools for:

  • Detection: Identifying objects within images.
  • Detailing: Enhancing the details of specific regions in images.
  • Upscaling: Increasing the resolution of images.
  • Pipes: Streamlining complex workflows.

ONNXDetectorProvider Introduction:

The ONNXDetectorProvider node allows users to load and utilize ONNX-based object detection models within ComfyUI. ONNX (Open Neural Network Exchange) is an open standard for representing machine learning models, enabling interoperability between different frameworks. This node provides a way to integrate pre-trained ONNX detection models into ComfyUI workflows.

ONNXDetectorProvider Input:

The ONNXDetectorProvider node accepts the following input:

  • model_name: (Dropdown List) - A selection of available ONNX models listed in the "onnx" folder within the ComfyUI models directory. The models listed will be the files found using folder_paths.get_filename_list("onnx").

ONNXDetectorProvider Output:

The ONNXDetectorProvider node outputs the following:

  • BBOX_DETECTOR: An object that represents the loaded ONNX object detection model. This output can be connected to other nodes that require a bounding box detector, such as ONNXDetectorForEach.

ONNXDetectorProvider Usage Tips:

  • Model Placement: Ensure that your ONNX models are placed in the correct directory (typically ComfyUI/models/onnx) so that they appear in the model_name dropdown list. The code explicitly uses os.path.join(model_path, "onnx") and adds the .onnx extension.
  • Dependency: This node relies on the impact.core.ONNXDetector class. Ensure that the Impact Pack is correctly installed and that the necessary dependencies for ONNX inference are met.
  • Versatility: The BBOX_DETECTOR output can be used with various downstream nodes within the Impact Pack to perform tasks such as object detection, segmentation, and detailing on specific regions of an image.
  • Integration: The ONNXDetectorProvider is often used in conjunction with nodes like ONNXDetectorForEach to apply the detection model to an image and process the detected regions.

related extension: