def getSubSample(self, idx, normImgPatch):
imgPatch = self.imgData[:, :, idx[0]:idx[0] + idx[3],
idx[1]:idx[1] + idx[4], idx[2]:idx[2] + idx[5]]
labelData = torch.empty(
(1, self.imgData.shape[1], idx[3], idx[4], idx[5]),
requires_grad=False)
if normImgPatch:
imgPatch = normalizeImg(imgPatch)
if (self.labelData.dim() == self.imgData.dim()):
labelData = self.labelData[:, :, idx[0]:idx[0] + idx[3],
idx[1]:idx[1] + idx[4],
idx[2]:idx[2] + idx[5]]
return (imgPatch, labelData)
def getRandomSubSamples(self,
numberofSamplesPerRun,
idxs,
normImgPatch=False):
imgDataNew = torch.empty(
(numberofSamplesPerRun, self.imgData.shape[1], self.patchSizes[0],
self.patchSizes[1], self.patchSizes[2]),
requires_grad=False)
if (self.labelData.dim() == self.imgData.dim()):
labelDataNew = torch.empty(
(numberofSamplesPerRun, self.imgData.shape[1],
self.patchSizes[0], self.patchSizes[1], self.patchSizes[2]),
requires_grad=False)
randSampleIdxs = np.random.randint(0, len(idxs[0]),
(numberofSamplesPerRun, ))
usedIdx = []
for j in range(0, numberofSamplesPerRun):
idx0 = idxs[0][randSampleIdxs[j]]
idx2 = idxs[1][randSampleIdxs[j]]
idx3 = idxs[2][randSampleIdxs[j]]
idx4 = idxs[3][randSampleIdxs[j]]
usedIdx.append((idx2, idx3, idx4, self.patchSizes[0],
self.patchSizes[1], self.patchSizes[2]))
imgPatch = self.imgData[idx0, :, idx2:idx2 + self.patchSizes[0],
idx3:idx3 + self.patchSizes[1],
idx4:idx4 + self.patchSizes[2]]
if normImgPatch:
imgPatch = normalizeImg(imgPatch)
imgDataNew[j, ] = imgPatch
# indexArrayTest[idx2:idx2 + patchSizes[0], idx3:idx3 + patchSizes[1], idx4:idx4 + patchSizes[2]] += 1
if (self.labelData.dim() == self.imgData.dim()):
labelDataNew[j, ] = self.labelData[idx0, :, idx2:idx2 +
self.patchSizes[0],
idx3:idx3 +
self.patchSizes[1],
idx4:idx4 +
self.patchSizes[2]]
imgDataToWork = imgDataNew
if (self.labelData.dim() == self.imgData.dim()):
labelDataToWork = labelDataNew
else:
labelDataToWork = torch.Tensor()
return (imgDataToWork, labelDataToWork, usedIdx)
def getUniformlyDistributedSubsamples(self,
numberofSamplesPerRun,
idxs,
currIteration,
normImgPatch=False):
startIdx = currIteration % numberofSamplesPerRun
imgDataNew = torch.empty(
(numberofSamplesPerRun, self.imgData.shape[1], self.patchSizes[0],
self.patchSizes[1], self.patchSizes[2]),
requires_grad=False)
if (self.labelData.dim() == self.imgData.dim()):
labelDataNew = torch.empty(
(numberofSamplesPerRun, self.imgData.shape[1],
self.patchSizes[0], self.patchSizes[1], self.patchSizes[2]),
requires_grad=False)
iterationRange = np.arange(startIdx, startIdx + numberofSamplesPerRun)
iterationRange[iterationRange >= len(
idxs)] = iterationRange[iterationRange >= len(idxs)] - len(idxs)
j = 0
for i in iterationRange:
idx = idxs[i]
imgPatch = self.imgData[:, :, idx[0]:idx[0] + idx[3],
idx[1]:idx[1] + idx[4],
idx[2]:idx[2] + idx[5]]
if normImgPatch:
imgPatch = normalizeImg(imgPatch)
imgDataNew[j, ] = imgPatch
if (self.labelData.dim() == self.imgData.dim()):
labelDataNew[j, ] = self.labelData[:, :,
idx[0]:idx[0] + idx[3],
idx[1]:idx[1] + idx[4],
idx[2]:idx[2] + idx[5]]
j = j + 1
imgDataToWork = imgDataNew
if (self.labelData.dim() == self.imgData.dim()):
labelDataToWork = labelDataNew
else:
labelDataToWork = torch.Tensor()
return (imgDataToWork, labelDataToWork)
def getSubSampleImg(self, idx, normImgPatch):
imgPatch = self.imgData[:, :, idx[0]:idx[0] + idx[3],
idx[1]:idx[1] + idx[4], idx[2]:idx[2] + idx[5]]
if normImgPatch:
imgPatch = normalizeImg(imgPatch)
return imgPatch