search

WXinlong / SOLO

SOLO and SOLOv2 for instance segmentation, ECCV 2020 & NeurIPS 2020.
Other
1.69k stars 307 forks source link

The inference speed is slower when upgrade to pytorch>1.5 #176

Open elepherai opened 3 years ago

elepherai commented 3 years ago

I pull the latest codes from c7b294a , update base image to pytorch/pytorch:1.8.1-cuda11.1-cudnn8-devel.

Found that the inference speed is 30% slower than before: pytorch/pytorch:1.3-cuda10.1-cudnn7-devel.

1.6.0-cuda10.1-cudnn7-devel and 1.8.1-cuda10.2-cudnn7-devel are also slower.