지속 가능한 꾸준함

PyTorch 모델을 ONNX 형식으로 변환할 때, 모델의 forward 메소드가 추가 인자를 받는 경우가 있습니다. 예를 들어, 학습/추론 모드를 구분하는 phase 인자와 같은 것들이죠. 이런 경우 ONNX 변환 시 어떻게 처리해야 하는지 알아보겠습니다. 문제 상황 일반적으로 PyTorch 모델의 forward 메소드는 입력 텐서만 받지만, 때로는 다음과 같이 추가 인자를 받도록 구현되어 있을 수 있습니다:def forward(self, x, phase='train'): # phase에 따라 다른 동작을 수행 if phase == 'train': # 학습 시의 동작 ... else: # 추론 시의 동작 ... return outp..

for env in $(conda env list | grep -v '^#' | awk '{print $1}'); do conda env remove -n "$env" -y; done

https://arxiv.org/pdf/2003.10168 https://arxiv.org/pdf/2102.05447 https://arxiv.org/pdf/2407.14972   https://github.com/serengil/deepface/tree/master/benchmarks deepface/benchmarks at master · serengil/deepfaceA Lightweight Face Recognition and Facial Attribute Analysis (Age, Gender, Emotion and Race) Library for Python - serengil/deepfacegithub.com   face recognition 관련해서 찾다보면 embedding feature..