EthanH3514/AL_Yolo

What it solves

AL-YOLO is designed for real-time visual target detection and tracking, specifically optimized for high-speed environments like gaming scenarios. It addresses the problemas of input latency and processing overhead that typically hinder real-time computer vision systems.

How it works

The system implements an end-to-end pipeline that integrates several key components:

1. Screen Capture: Uses DXShot to achieve high-speed screen grabbing (reducing capture time from 15ms to 5ms).
2. Image Processing: Utilizes OpenCV for pre-processing images before they are fed into the model.
3. Inference: Employs a YOLOv5s model running on PyTorch and CUDA for GPU-accelerated target detection.
4. Target Detection: Processes the model output to identify targets and determine their priority.
5. Control Interface: Provides a lightweight GUI for parameter configuration and real-time visualization of detection results.

Who it’s for

Developers and researchers interested in real-time computer vision, GPU inference, and low-latency visual processing, particularly those experimenting with target detection in dynamic, high-frequency environments.

Highlights

• High Performance: Achieves approximately 50 FPS on an NVIDIA RTX 3050 Laptop GPU.
• Low Latency: Optimized screen capture via DXShot to minimize input delay.
• End-to-End Pipeline: Integrates capture, detection, and visualization in a single workflow.
• Adaptive Parameters: Supports parameter adaptation across different machine environments.
• Priority-Based Detection: Capable of multi-target recognition and priority judgment.