search

ChenhongyiYang / QueryDet-PyTorch

[CVPR 2022 Oral] QueryDet: Cascaded Sparse Query for Accelerating High-Resolution Small Object Detection
MIT License
432 stars 47 forks source link

Does the environment configuration affect the experimental results? #96

Closed liuw20 closed 6 months ago

liuw20 commented 6 months ago

my environment： RTX 3090+Pytorch1.12.0+cu113+detectron0.6

my result: image

I trained retinanet and querydet with the default config，but the results are not same with the paper. Is it because of environmental issues?

liuw20 commented 6 months ago

visdrone+2019

liuw20 commented 6 months ago

I adjust the version of spconv to 'spconv-cu113==2.1.25'，the result is perfect! So the version of spconv must be matched with the cuda version!

judycpChen commented 6 months ago

I adjust the version of spconv to 'spconv-cu113==2.1.25'，the result is perfect! So the version of spconv must be matched with the cuda version!

Hello ,do you have this problem? NotImplementedError: SPCONV_FILTER_HWIO is deprecated. use SPCONV_SAVED_WEIGHT_LAYOUT instead.

judycpChen commented 5 months ago

I adjust the version of spconv to 'spconv-cu113==2.1.25'，the result is perfect! So the version of spconv must be matched with the cuda version!

Hello, I would like to know how your configuration is, your platform is windows or other