Thank you for open-sourcing your code. When I run validation test using your pretrained weight, the result is quite low for polar version: mIoU = 32.208 with car's mIoU=87.51. Meanwhile, for traditional pretrained weight the result if mIoU = 37.412 with car's mIoU = 89.01. This is quite different from the reported result. I was wondering if this is caused by higher version of CUDA & Pytorch? I am using Cuda11.1 with Pytorch 1.8.1. Have you tested with this? I am looking forward to your reply. Thank you in advance.
Hi @edwardzhou130,
Thank you for open-sourcing your code. When I run validation test using your pretrained weight, the result is quite low for polar version: mIoU = 32.208 with car's mIoU=87.51. Meanwhile, for traditional pretrained weight the result if mIoU = 37.412 with car's mIoU = 89.01. This is quite different from the reported result. I was wondering if this is caused by higher version of CUDA & Pytorch? I am using Cuda11.1 with Pytorch 1.8.1. Have you tested with this? I am looking forward to your reply. Thank you in advance.
Best,