search

haofeixu / aanet

[CVPR'20] AANet: Adaptive Aggregation Network for Efficient Stereo Matching
Apache License 2.0
521 stars 100 forks source link

About Correlation layer and can you share me the code without deformable conv #57

Closed yanzhou-li closed 3 years ago

yanzhou-li commented 3 years ago

i come back again. Drive into the theme,I want to know if I use the concatand convolution instead of correlation,the net can work? if it works, how can I do that,thanks! @haofeixu @Ilya-Muromets

haofeixu commented 3 years ago

@yanzhou-li , yes, our implementation supports both cost volume construction methods, you can check the --feature_similarity argument in train.py.

haofeixu commented 3 years ago

BTW, if you don't want to use the deformable conv, you can set the --num_deform_blocks to 0 in train.py.

yanzhou-li commented 3 years ago

@haofeixu i got it. thank you