search

haofeixu / aanet

[CVPR'20] AANet: Adaptive Aggregation Network for Efficient Stereo Matching
Apache License 2.0
534 stars 102 forks source link

FLOPs 是如何得出的 #16

Closed JumpXing closed 4 years ago

JumpXing commented 4 years ago

请问下你们的各个模型的FLOPs 是如何得出的？ 是只计算了CONV中的MAC次数，还是计算量CONV、BN、activate function、pooling等所有的

haofeixu commented 4 years ago

The FLOPs is measured for conv layers only and you can refer to this repo: https://github.com/Lyken17/pytorch-OpCounter

JumpXing commented 4 years ago

Thanks for your explaining. We also used this tool but got a different result of FLOPs from yours. Now I know that reason is that our result contains not only CONV but also BN, pooling and so on.

lelelexxx commented 4 years ago

Thanks for your explaining. We also used this tool but got a different result of FLOPs from yours. Now I know that reason is that our result contains not only CONV but also BN, pooling and so on.

Hi! I'm also curious about the FLOPs of network, Could you please tell me which libs or tools did you use for FLOPs calculation? Thanks a lot!