Closed lucasjinreal closed 2 years ago
150ms average, is that normal? I changed everything into GPU.
using r50 version
We spend about 140ms per image on 4 Tesla V100. We find that the inference time using multiple GPUs is relatively stable. Moreover, the depth of transformer has a great influence on inference time.
150ms average, is that normal? I changed everything into GPU.
using r50 version