Runrazor Detectors

Runrazor Detectors provides downloadable object detector model weights designed for use with the runrazor tracking system. The models are based on stock YOLOX object detectors and are exported to the ONNX format. These detectors are specifically configured to operate on the COCO person class, meaning they detect people in video frames, including scenarios where skiers are identified as people. The tracking and subject selection beyond basic detection are managed by runrazor's own tracking layer. The available models include variants such as yolox_s-v1.onnx, which offers a balance between accuracy and speed and is recommended as an upgrade, and yolox_m-v1.onnx, which provides higher recall at the cost of slower performance on CPUs. The smaller yolox_nano model is bundled directly with the runrazor package and does not require separate downloading. When a user specifies a model such as yolox_s for use with runrazor, the system fetches the model file on demand, verifies its integrity using SHA-256, and caches it locally for future use. Each model file comes pre-stamped with runrazor-specific runtime settings in the ONNX metadata, including the format, confidence threshold, and class ID, allowing for seamless integration without the need for additional configuration flags. The detector weights are distributed under the Apache 2.0 license. The tool is positioned for users of the runrazor tracking system who require person detection capabilities in their video analysis workflows. No further details about broader features, integrations, or additional supported classes are provided in the available evidence.