is it the correct version?

[Zealoe]

is it the correct version of deformable conv? and as for the training details, do you have some advices? for examples,learning rate？

[Zardinality]

I guess so, you can compare the result with the original version deform conv in mxnet. I have tested it with some random input. I would be appreciated if you find some counter examples. Although the cuda code are the same, the tensorflow wrap is kind of messy so I do not make 100 percent guarantee. I remember the original author made some comments about the training. But I don't have much experiences tuning it, sorry.