AI Models Used (YOLOv8, PyTorch, TensorRT)

Role in the Project

The AI models in EYNTRY are responsible for object detection and recognition, ensuring high accuracy and real-time processing. The system utilizes YOLOv8 for detection, PyTorch for training, and TensorRT for optimized inference on edge devices.

Strengths & Weaknesses

Strengths:

• YOLOv8: Real-time processing with high precision.
• PyTorch: Strong research community and flexibility for custom training.
• TensorRT: Optimized inference for low-latency AI applications.

Weaknesses:

• YOLOv8: Requires tuning for different datasets.
• PyTorch: Training can be computationally expensive.
• TensorRT: Limited support for dynamic models.

Available Technologies & Comparison

• YOLOv8 (Chosen) vs. SSD (Lower accuracy) vs. Faster R-CNN (Slower, more precise).
• PyTorch (Chosen) vs. TensorFlow (Better for deployment, less flexibility for research).
• TensorRT (Chosen for Edge AI) vs. ONNX Runtime (More general, less optimized for NVIDIA GPUs).

Chosen Approach

• YOLOv8 for real-time object detection.
• PyTorch for flexible model training.
• TensorRT for efficient deployment on edge devices.

Example of YOLOv8 inference with PyTorch:

from ultralytics import YOLO
model = YOLO('yolov8n.pt')
results = model.predict(source='image.jpg', conf=0.5)
print(results)