# 推理
onnxruntime-directml>=1.18
numpy>=1.26

# 训练 / 导出
torch>=2.6.0
opencv-python>=4.9
onnx>=1.16
onnxconverter-common>=1.16