Is it necessary to build deformable conv?

haofeixu / aanet

[CVPR'20] AANet: Adaptive Aggregation Network for Efficient Stereo Matching

Apache License 2.0

524 stars 102 forks source link

Is it necessary to build deformable conv? #24

Closed xuzhongyou closed 4 years ago

xuzhongyou commented 4 years ago

The new pytorch version has support deformable conv

haofeixu commented 4 years ago

We recommend to build deformable conv on your own, since it's very easy and I haven't used the deformable conv implementation by pytorch, you may need to configue yourself and the pretrained weights may not be compatible.