inference speed is much slower than as stated

[Libaishun]

Hi, thank you for the work ! I have test on my own segmentation task, with image input size 15361536, the inference speed on my 1080ti is about 300ms per image, which is far slower than 53fps with input size 10242048 as stated. Have you used tensorrt or any acceleration method ?

[PingoLH]

We use native pytorch 1.0.1 with cuda 9.2 for the speed test. You should be able to reproduce a similar speed on 0.4.0 or versions above 1.2.0 with cuda 10.1. The actual processing frame rate will be lower than the reported due to file IO, so in the validate.py we count inference time only for GPU. Also, using save_image can significantly slow down the processing speed. Please check your file system speed if you still have this problem.

[purse1996]

Thank you. I test the code in the 1080TI. The inference time is very fast, but the file IO is very very slow. Can you offer some suggestions about improving file IO speed?