Pytorch_Retinaface
Pytorch_Retinaface copied to clipboard
Really slow on GPU
trafficstars
input a image 1920*1080 , cost 0.11~0.14s , on 2070S .
I use torchvision.io.read_image (torchvision >= 0.8.0) function to load image data as torch.tensor, then put it on GPU for data preprocessing, it can reduce the time consuming of data preprocessing on the CPU.
https://github.com/xxcheng0708/Pytorch_Retinaface_Accelerate