Hi, based on my understanding, receptive field size is same as the convolutional filter size, according to model initialization under the training code, filter size is 3x3 for all the layers, so receptive field size is 3x3 for all layers regardless the depth of the network.
However, the report says: "the receptive field of DnCNN with depth of d should be (2d + 1) x (2d + 1). Thus, for Gaussian denoising with a certain noise level, we set the receptive field size of DnCNN to 35x35 with the corresponding depth of 17."
Do you mind further explain it so that I know where I am wrong?
Thanks!
Hi, based on my understanding, receptive field size is same as the convolutional filter size, according to model initialization under the training code, filter size is 3x3 for all the layers, so receptive field size is 3x3 for all layers regardless the depth of the network. However, the report says: "the receptive field of DnCNN with depth of d should be (2d + 1) x (2d + 1). Thus, for Gaussian denoising with a certain noise level, we set the receptive field size of DnCNN to 35x35 with the corresponding depth of 17." Do you mind further explain it so that I know where I am wrong? Thanks!