Simple Question about offset & dec dimension

[hangil55]

Thank you for caffe version of DCN. I have some questions 1) why offset & dec dimmension is 72 dim, 512 dim? I think offset is eaqual to bottom dimemsion.... 2) why dec layer is needed? 3) In Resnet 50 proto, you apply RCNN layer. however, DCN is able to apply all the convolution layer (e.g., VGG16, conv1 ~ 5) 4) what is the role the layer of offset & dec

thank you

[unsky]

@hangil55 i dont understand what your mean. The params in DeformableConvolution: bottom0: (batch_size, channel, height, width) bottom[1] (offset): (batch_size, deformable_group kernel[0] kernel[1]*2, height, width) you can read the paper in deformable conv to understand why..

dec layer is core of DeformableConvolution, why you want to throw it?

[hangil55]

Thank you for fast reply.. I think 72 dimension in your offset parameters. how calculate "(offset): (batch_size, deformable_group kernel[0] kernel[1]*2, height, width)" dimensions?

[lawpdas]

What is the meaning of every channel in offset? For each position: [x1, x2 ... xn, y1, y2, y3 ... yn] or [x1, y1, x2, y2 ... xn, yn]?