In model inference scenarios, developers often need to invoke visual recognition capabilities to process images. xby-detect is a DSH plugin maintained by xby-skill, providing a complete set of visual detection and recognition capabilities.

Introduction

xby-detect belongs to the model inference category and is licensed under the MIT License. It integrates multiple visual tasks, supporting recognition from everyday object detection to specific scenarios (such as flame and reflective vest detection), covering multiple domains including insects, plants, animals, and vehicles.

Installation and Activation

Install the plugin in a DSH environment by running the following command:

dsh plugin --profile web add github:xby_skill/xby-detect

After installation, set an API key in the conversation to enable detection features:

set_xby_apikey

Core Features

The plugin offers a rich set of detection and recognition commands, mainly divided into the following categories:

  • General and Specialized Detection
    Includes everyday object detection (daily_object_detection), general recognition (general_recognition), vehicle detection (detect_vehicle), pedestrian detection (detect_pedestrian), reflective vest detection (detect_reflective_vest), safety helmet and person detection (helmet_head_person_detection), etc.

  • Domain-Specific Recognition
    Covers insect recognition (insect_recognition), plant recognition (plant_recognition), animal recognition (animal_recognition), bird recognition (bird_detection), wild animal detection (wild_animals_detection), pet emotion recognition (pet_emotion_recognition), dish recognition (dish_recognition), etc.

  • Scenario and Behavior Detection
    Includes cellphone detection (cellphone_detection), electric bike detection (ebike_detection), gesture detection (gesture_detection), smoking detection (smoking_detection), and flame detection (fire_detection).

  • YOLO26 Model Series
    Provides a variety of visual tasks based on the YOLO26 series:

    • Object detection: yolo26_detect (80 COCO classes)
    • Oriented bounding box detection: yolo26_obb_detect (15 specific classes)
    • Instance segmentation: yolo26_segment (80 COCO classes)
    • Human pose estimation: yolo26_pose_estimate (17 keypoints)
    • Image classification: yolo26_classify (1000 ImageNet classes)

Typical Usage

Taking a BASE64-encoded input image as an example, invoke the everyday object detection feature:

daily_object_detection_for_data_base64

The system outputs the detection boxes, confidence scores, and labels for all person/pet/vehicle/flame/cardboard box targets in the image.

Invoke the pet emotion recognition feature:

pet_emotion_recognition

The system recognizes the pet’s facial expression and outputs one of four emotions: Angry / Happy / Relaxed / Sad.

Notes

  1. Input format: Most detection features support three formats: image file links, BASE64 encoding, or file paths.
  2. API key: You must set an API key using the set_xby_apikey command before use.
  3. Scenario suitability: Features such as flame detection perform best from security camera and traffic camera viewpoints.

Conclusion

xby-detect provides foundational and rich visual recognition capabilities for the DSH plugin ecosystem. After installation and configuration, developers can invoke various detection interfaces in conversations or scripts.