(PPS) Deformable Convolutional Networks
The basic idea of this paper is to give the convolution and pooling layers the ability to model different orientations and scales of objects in images. They do this by making the shape of the convolution filter learnable. For intuition, consider the…