tusen-ai / SST

Code for a series of work in LiDAR perception, including SST (CVPR 22), FSD (NeurIPS 22), FSD++ (TPAMI 23), FSDv2, and CTRL (ICCV 23, oral).
Apache License 2.0
801 stars 102 forks source link

About measuring model compelexities #80

Closed seonhoon1002 closed 2 years ago

seonhoon1002 commented 2 years ago

Hello, Can I ask you how to measure the model complexities of #param and Memory in the SST paper?

Abyssaledge commented 2 years ago

The number of parameters is calculated in a handcrafted way. As for memory, we report the memory reported in the log at the begining of second epoch.

seonhoon1002 commented 2 years ago

So in the memory report, did it measure in inference time?

Abyssaledge commented 2 years ago

No, we report the memory footprint in training time.

seonhoon1002 commented 2 years ago

Ok, I understand. Thanks!