blob: 3a75b3f7d8dad6f9532891bf1e7743f926534986 [file] [log] [blame]
#-------------------------------------------------------------
#
# Licensed to the Apache Software Foundation (ASF) under one
# or more contributor license agreements. See the NOTICE file
# distributed with this work for additional information
# regarding copyright ownership. The ASF licenses this file
# to you under the Apache License, Version 2.0 (the
# "License"); you may not use this file except in compliance
# with the License. You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing,
# software distributed under the License is distributed on an
# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
# KIND, either express or implied. See the License for the
# specific language governing permissions and limitations
# under the License.
#
#-------------------------------------------------------------
args <- commandArgs(TRUE)
library("Matrix")
imgSize=as.integer(args[1])
numImg=as.integer(args[2])
numChannels=as.integer(args[3])
numFilters=as.integer(args[4])
filterSize=as.integer(args[5])
stride=as.integer(args[6])
pad=as.integer(args[7])
P=as.integer(args[8])
Q=as.integer(args[9])
# Assumption: NCHW image format
w=matrix(seq(1, numFilters*numChannels*filterSize*filterSize), numFilters, numChannels*filterSize*filterSize, byrow=TRUE)
dout=matrix(seq(1, numImg*numFilters*P*Q), numImg, numFilters*P*Q, byrow=TRUE)
if(as.logical(args[11])) {
zero_mask = (w - mean(w)) > 0
w = w * zero_mask
} else {
w = w - mean(w)
}
if(as.logical(args[12])) {
zero_mask = (dout - mean(dout)) > 0
dout = dout * zero_mask
} else {
dout = dout - mean(dout)
}
col2im <- function(img_cols, C, Hin, Win, Hf, Wf,
strideh, stridew, reduction) {
Hout = as.integer((Hin - Hf) / strideh + 1)
Wout = as.integer((Win - Wf) / stridew + 1)
img = matrix(0, C, Hin*Win, byrow=TRUE) # zeros
for (hout in 1:Hout) { # all output rows
hin = (hout-1) * strideh + 1
for (wout in 1:Wout) { # all output columns
win = (wout-1) * stridew + 1
# Extract a local patch of the input image corresponding spatially to the filter sizes.
img_patch = matrix(img_cols[,(hout-1)*Wout + wout], C, Hf*Wf, byrow=TRUE) # zeros
for (c in 1:C) { # all channels
img_patch_slice = matrix(img_patch[c,], Hf, Wf, byrow=TRUE) # reshape
if (reduction == "add") {
img_slice = matrix(0, Hin, Win, byrow=TRUE)
img_slice[hin:(hin+Hf-1), win:(win+Wf-1)] = img_patch_slice
img[c,] = img[c,] + matrix(t(img_slice), 1, Hin*Win)
} else {
img_slice = matrix(img[c,], Hin, Win, byrow=TRUE)
img_slice[hin:(hin+Hf-1), win:(win+Wf-1)] = img_patch_slice
img[c,] = matrix(t(img_slice), 1, Hin*Win)
}
}
}
}
img
}
unpad_image <- function(img_padded, Hin, Win, padh, padw) {
C = nrow(img_padded)
img = matrix(0, C, Hin*Win, byrow=TRUE)
for (c in 1:C) {
img_padded_slice = matrix(img_padded[c,], (Hin+2*padh), (Win+2*padw), byrow=TRUE)
img_slice = img_padded_slice[(padh+1):(padh+Hin), (padw+1):(padw+Win)]
img[c,] = matrix(t(img_slice), 1, Hin*Win)
}
img
}
conv2d_backward_data <- function(dout, Hout, Wout,
W, N, C, Hin, Win, Hf, Wf,
strideh, stridew, padh, padw) {
F = nrow(W)
# Create gradient volumes
dX = matrix(0, N, C*Hin*Win, byrow=TRUE)
# Partial derivatives for convolution - im2col implementation
for (n in 1:N) { # all examples
doutn = matrix(dout[n,], F, Hout*Wout, byrow=TRUE)
# Compute dX
dXn_padded_cols = t(W) %*% doutn # shape (C*Hf*Wf, Hout*Wout)
dXn_padded = col2im(dXn_padded_cols, C, Hin+2*padh, Win+2*padw, Hf, Wf, strideh, stridew, "add")
dXn = unpad_image(dXn_padded, Hin, Win, padh, padw)
dX[n,] = matrix(t(dXn), 1, C*Hin*Win) # reshape
}
dX
}
dx = conv2d_backward_data(dout, P, Q, w, numImg, numChannels, imgSize, imgSize, filterSize, filterSize, stride, stride, pad, pad);
writeMM(as(dx,"CsparseMatrix"), paste(args[10], "B", sep=""))