hanchaoleng / ShapeConv

ShapeConv: Shape-aware Convolutional Layer for Indoor RGB-D Semantic Segmentation (ICCV 2021)
Apache License 2.0
106 stars 15 forks source link

Scales for Test #5

Closed Zongwei97 closed 2 years ago

Zongwei97 commented 2 years ago

Hello,

Thanks for the great work !

I am confused about the scales in the configs. Why you use multi-scales for testing instead of scale = 1?

For example the code on the configs/nyu/nyu40 tta=dict( scales=[0.5, 0.75, 1.0, 1.25, 1.5, 1.75],

Can you provide some explications? Thanks

hanchaoleng commented 2 years ago

Multi-scale testing achieve better performance while consuming more computing power.If you want to test on single-scale, you should comment out the tta.We report both single-scale and multi-scale results in our paper and compare them with previous work on the same testing strategy.

Zongwei97 commented 2 years ago

Thanks for reply !