Closed fushh closed 1 year ago
Till now, we don't have a plan to release our customized mmcv as it is not systematic and well structured. You can compute extra FLOPs according to SRShadowForFlops
manually or resort to fvcore
(a simple but comprehensive FLOP statistics tool, better than mmcv).
When using officially downloaded mmcv-full=1.3.3
, it raises many import errors like
ImportError: cannot import name 'to_2tuple' from 'mmcv.utils' (/home/proto/anaconda3/envs/mmcv133/lib/python3.7/site-packages/mmcv/utils/init.py) ImportError: cannot import name 'trunc_normal_init' from 'mmcv.cnn' (/home/proto/anaconda3/envs/mmcv133/lib/python3.7/site-packages/mmcv/cnn/init.py)
Does these NotImplementedErrors come from your customized mmcv
?
I use the
tools/analysis_tools/get_flops.py
to calculate AdaMixer's FLOPs and get the following results which is not aligned with the number reported in the paper. And it seems that theSRShadowForFlops
doesn't work.Could you tell me how to get the exact FLOPs number?