The different reported performances between table 1 and table 4.

speedinghzl / CCNet

CCNet: Criss-Cross Attention for Semantic Segmentation (TPAMI 2020 & ICCV 2019).

MIT License

1.42k stars 277 forks source link

The different reported performances between table 1 and table 4. #47

Closed tanxin2017 closed 4 years ago

tanxin2017 commented 4 years ago

Hi, Thanks for you work. I am curious that why the preformance of CCNET on table 1 is 81.3 while that on table 4 is 79.8. It seems they are under the same setting and should be the same network. Do I miss some details?

Thanks!

speedinghzl commented 4 years ago

In Table 1, we use multi-scale and horizontal flip strategies for inference. In Table 4, we do not take these strategies.

tanxin2017 commented 4 years ago

Thanks!