diff --git a/UNet/NormalizeTrainingdata_32.ipynb b/UNet/NormalizeTrainingdata_32.ipynb
index 90d947de6d171fa679344098864bb6545f48a818..507145720e2b0586e42a31811f166d19c6ec202f 100644
--- a/UNet/NormalizeTrainingdata_32.ipynb
+++ b/UNet/NormalizeTrainingdata_32.ipynb
@@ -2,7 +2,7 @@
"cells": [
{
"cell_type": "code",
- "execution_count": null,
+ "execution_count": 1,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
@@ -10,34 +10,24 @@
"id": "-Gdj75rhq3Pe",
"outputId": "57ff2b53-db05-4526-8d22-98dbdf52fd56"
},
- "outputs": [
- {
- "name": "stdout",
- "output_type": "stream",
- "text": [
- "Drive already mounted at /content/drive; to attempt to forcibly remount, call drive.mount(\"/content/drive\", force_remount=True).\n"
- ]
- }
- ],
+ "outputs": [],
"source": [
"import torch\n",
"import numpy as np\n",
"import torch.nn as nn\n",
- "from torch.utils.data import TensorDataset\n",
- "from google.colab import drive\n",
- "drive.mount('/content/drive')"
+ "from torch.utils.data import TensorDataset"
]
},
{
"cell_type": "code",
- "execution_count": null,
+ "execution_count": 4,
"metadata": {
"id": "OzNQI96lq3Pi"
},
"outputs": [],
"source": [
- "training_data = np.load('/content/drive/MyDrive/Bachlorarbeit/Trainingsdata/Training_data.npy')\n",
- "training_label = np.load('/content/drive/MyDrive/Bachlorarbeit/Trainingsdata/Training_labels.npy')\n",
+ "training_data = np.load('E:/Data/damask3/UNet/Input/Training_data_32.npy')\n",
+ "training_label = np.load('E:/Data/damask3/UNet/Input/Training_labels_32.npy')\n",
" #Desired input shape: (N,C,D,H,W) N=Batchsize, C=Channel, D=Depth, H=Height, W = Width\n",
"data = np.moveaxis(training_data,4,1) # GAN output has shape(N,W,H,D,C) but no need to change spatial dimensions, but Channel dimension must be changed\n",
"#input[0-3]: Orientation, input[4]: Phase(one for martinsite)"
@@ -69,7 +59,7 @@
},
{
"cell_type": "code",
- "execution_count": null,
+ "execution_count": 6,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
@@ -82,7 +72,7 @@
"name": "stdout",
"output_type": "stream",
"text": [
- "size of input is torch.Size([1987, 6, 32, 32, 32])\n",
+ "size of input is torch.Size([1987, 2, 32, 32, 32])\n",
"size of label is torch.Size([1987, 32, 32, 32])\n"
]
}
@@ -93,8 +83,9 @@
"phase= data[:,4,:,:,:].reshape(1987, 1,32,32,32)\n",
"new_phase = np.ones(phase.shape) - phase #input[4]: martinsite, input[5]:ferrit\n",
"#new_training_data = np.append(data,new_channel,axis=1)\n",
- "input = np.append(angles,phase,axis=1)\n",
- "input = np.append(input,new_phase,axis=1)\n",
+ "#input = np.append(angles,phase,axis=1)\n",
+ "#input = np.append(input,new_phase,axis=1)\n",
+ "input = np.append(phase,new_phase,axis=1)\n",
"\n",
"label = torch.from_numpy(training_label)\n",
"input = torch.from_numpy(input)\n",
@@ -110,7 +101,7 @@
},
{
"cell_type": "code",
- "execution_count": null,
+ "execution_count": 7,
"metadata": {
"id": "-Rbt8Brb9mM_"
},
@@ -128,7 +119,7 @@
},
{
"cell_type": "code",
- "execution_count": null,
+ "execution_count": 9,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
@@ -141,17 +132,17 @@
"name": "stdout",
"output_type": "stream",
"text": [
- "tensor(0., dtype=torch.float64)\n"
+ "tensor(1., dtype=torch.float64)\n"
]
}
],
"source": [
- "print(input.min())"
+ "print(input.max())"
]
},
{
"cell_type": "code",
- "execution_count": null,
+ "execution_count": 11,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
@@ -159,20 +150,13 @@
"id": "W9oIMYy89kXY",
"outputId": "950ee423-0a95-43d7-8fee-9f71d960a5dc"
},
- "outputs": [
- {
- "name": "stderr",
- "output_type": "stream",
- "text": [
- "/usr/local/lib/python3.7/dist-packages/numpy/core/_asarray.py:136: VisibleDeprecationWarning: Creating an ndarray from ragged nested sequences (which is a list-or-tuple of lists-or-tuples-or ndarrays with different lengths or shapes) is deprecated. If you meant to do this, you must specify 'dtype=object' when creating the ndarray\n",
- " return array(a, dtype, copy=False, order=order, subok=True)\n"
- ]
- }
- ],
+ "outputs": [],
"source": [
"dataset = TensorDataset(input,label_normalized) # create the pytorch dataset \n",
- "np.save('/content/drive/MyDrive/Bachlorarbeit/Trainingsdata/Norm_min_max_V2.npy',[min_label, max_label,angles_min_max])\n",
- "torch.save(dataset,'/content/drive/MyDrive/Bachlorarbeit/Trainingsdata/Training_Dataset_normalized_V2.pt')\n"
+ "#np.save('E:/Data/damask3/UNet/Input/Norm_min_max_32_V2.npy',[min_label, max_label,angles_min_max])\n",
+ "np.save('E:/Data/damask3/UNet/Input/Norm_min_max_32_V2.npy',[min_label, max_label])\n",
+ "\n",
+ "torch.save(dataset,'E:/Data/damask3/UNet/Input/Training_Dataset_normalized__32_V2.pt')\n"
]
}
],
diff --git a/UNet/UNet_V9.py b/UNet/UNet_V9.py
new file mode 100644
index 0000000000000000000000000000000000000000..db51335bef9054feefc0ec97c268c9a182781f7e
--- /dev/null
+++ b/UNet/UNet_V9.py
@@ -0,0 +1,250 @@
+#like V6_2 but only the different phases as input
+"""UNet_V6.ipynb
+
+Automatically generated by Colaboratory.
+
+Original file is located at
+ https://colab.research.google.com/drive/1yvtk3lFo_x0ZiqtFdnR8jgcjPKy3nZA4
+"""
+
+import torch
+import torch.nn as nn
+import numpy as np
+import random
+from torch.utils.data.sampler import SubsetRandomSampler
+from torch.utils.data.dataloader import DataLoader
+from torch.utils.data import TensorDataset
+import torch.nn.functional as F
+from torch.utils.data import random_split
+from torch.nn.modules.activation import ReLU
+
+class depthwise_separable_conv(nn.Module):
+ def __init__(self, in_c, out_1_c, out_2_c, padding, kernel_size):
+ super(depthwise_separable_conv, self).__init__()
+ self.depthwise_1 = nn.Conv3d(in_c, in_c, kernel_size= kernel_size, padding=padding[0], groups=in_c, bias=True)
+ self.pointwise_1 = nn.Conv3d(in_c, out_1_c, kernel_size=1, bias=True)
+ self.batch_norm_1 = nn.BatchNorm3d(out_1_c)
+ self.relu = nn.ReLU()
+ self.depthwise_2 = nn.Conv3d(out_1_c, out_1_c, kernel_size= kernel_size, padding=padding[1], groups=out_1_c, bias=True)
+ self.pointwise_2 = nn.Conv3d(out_1_c, out_2_c, kernel_size=1, bias=True)
+ self.batch_norm_2 = nn.BatchNorm3d(out_2_c)
+ def forward(self, x):
+ x = self.batch_norm_1(self.relu(self.pointwise_1(self.depthwise_1(x))))
+ return self.batch_norm_2(self.relu(self.pointwise_2(self.depthwise_2(x))))
+
+class convolution_Layer(nn.Module):
+ def __init__(self, in_c, out_1_c, out_2_c, padding, kernel_size):
+ super(convolution_Layer, self).__init__()
+ self.conv_1 = nn.Conv3d(in_c, out_1_c, kernel_size= kernel_size, padding=padding[0], bias=True)
+ self.batch_norm_1 = nn.BatchNorm3d(out_1_c)
+ self.relu = nn.ReLU()
+ self.conv_2 = nn.Conv3d(out_1_c, out_2_c, kernel_size= kernel_size, padding=padding[1], bias=True)
+ self.batch_norm_2 = nn.BatchNorm3d(out_2_c)
+ def forward(self, x):
+ x = self.batch_norm_1(self.relu(self.conv_1(x)))
+ return self.batch_norm_2(self.relu(self.relu(self.conv_2(x))))
+
+class head_layer(nn.Module):
+ def __init__(self, in_c, out_c = 1, padding = "same"):
+ super(head_layer, self).__init__()
+ self.conv = nn.Conv3d(in_c, out_c, kernel_size=1, bias=True)
+ self.sig = nn.Sigmoid()
+ def forward(self, x):
+ return self.sig(self.conv(x)) #convolution
+ #return self.sig(self.pointwise(self.depthwise(x))) #convolution
+
+class Encoder(nn.Module):
+ def __init__(self,kernel_size, chs, padding=((0,"same"),("same","same"),("same","same"))):
+ super().__init__()
+ self.channels = chs
+ self.enc_blocks = nn.ModuleList([depthwise_separable_conv(chs[i][0], chs[i][1], chs[i][2], kernel_size=kernel_size, padding=padding[i]) for i in range(len(chs))])
+ self.pool = nn.MaxPool3d(kernel_size=2, stride=2)
+ #self.batch_norm = nn.ModuleList([nn.BatchNorm3d( chs[i][2]) for i in range(len(chs))])
+ self.periodic_upsample = nn.ReflectionPad3d(int((kernel_size-1)/2))
+
+
+ def forward(self, x):
+ ftrs = []
+ x = self.periodic_upsample(x)
+ for i in range(len(self.channels)):
+ ftrs.append(x)
+ x =self.enc_blocks[i](x)
+ #print(f'size of ftrs: {ftrs[i].size()}')
+ x = self.pool(x)
+ #print(f'size of x after pooling{x.size()}')
+ ftrs.append(x)
+ #print(f'size of ftrs: {ftrs[3].size()}')
+ #print(f'length of ftrs: {len(ftrs)}')
+ return ftrs
+
+class Decoder(nn.Module):
+ def __init__(self,kernel_size, chs_upsampling, chs_conv, padding=(("same","same"),("same","same"),("same","same"))):
+ super().__init__()
+ assert len(chs_conv) == len(chs_upsampling)
+ self.chs = chs_upsampling
+ self.upconvs = nn.ModuleList([nn.ConvTranspose3d(chs_upsampling[i], chs_upsampling[i], 2, 2) for i in range(len(chs_upsampling))])
+ self.dec_blocks = nn.ModuleList([depthwise_separable_conv(chs_conv[i][0], chs_conv[i][1], chs_conv[i][2], kernel_size=kernel_size, padding=padding[i]) for i in range(len(chs_conv))])
+ self.head = head_layer(chs_conv[-1][2])
+ def forward(self, x, encoder_features):
+ for i in range(len(self.chs)):
+ x = self.upconvs[i](x)
+ #print(f'size after upsampling: {x.size()}')
+ enc_ftrs = self.crop(encoder_features[i], x)
+ x = torch.cat([x, enc_ftrs], dim=1)
+ #print(f'size after cropping&cat: {x.size()}')
+
+ x = self.dec_blocks[i](x)
+ #print(f'size after convolution: {x.size()}')
+ x = self.head(x)
+ return x
+
+ def crop(self, tensor, target_tensor):
+ target_size = target_tensor.size()[2]
+ tensor_size = tensor.size()[2]
+ delta = tensor_size - target_size
+ delta = delta // 2
+ return tensor[:,:,delta:tensor_size-delta,delta:tensor_size-delta,delta:tensor_size-delta]
+
+class UNetBase(nn.Module):
+ def training_step(self, batch):
+ input, labels = batch
+ out = self(input) # Generate predictions
+ loss = F.l1_loss(out, labels) # Calculate loss
+ return loss
+
+ def validation_step(self, batch):
+ input, labels = batch
+ out = self(input) # Generate predictions
+ loss = F.l1_loss(out, labels) # Calculate loss
+ acc = accuracy(out.detach(), labels.detach()) # Calculate accuracy
+ return {'val_loss': loss.detach(), 'val_acc': acc}
+
+ def validation_epoch_end(self, outputs):
+ batch_losses = [x['val_loss'] for x in outputs]
+ epoch_loss = torch.stack(batch_losses).mean() # Combine losses
+ batch_accs = [x['val_acc'] for x in outputs]
+ epoch_acc = torch.stack(batch_accs).mean() # Combine accuracies
+ return {'val_loss': epoch_loss.item(), 'val_acc': epoch_acc.item()}
+
+ def epoch_end(self, epoch, result):
+ print("Epoch [{}], train_loss: {:.6f}, val_loss: {:.6f}, val_acc: {:.6f}".format(
+ epoch, result['train_loss'], result['val_loss'], result['val_acc']))
+
+def accuracy(outputs, labels, threshold = 0.05):
+ error = (abs(outputs - labels)/outputs)
+ right_predic = torch.sum(error < threshold)
+ percentage = ((right_predic/torch.numel(error))*100.)
+ return percentage
+
+class UNet(UNetBase):
+ def __init__(self,kernel_size = 5, enc_chs=((2,16,32), (32,32,64), (64,64,128)), dec_chs_up=(128, 128, 64), dec_chs_conv=((192,128, 128),(160,64,64),(66,32,32))):
+ super().__init__()
+ self.encoder = Encoder(kernel_size = kernel_size, chs = enc_chs)
+ self.decoder = Decoder(kernel_size = kernel_size, chs_upsampling = dec_chs_up, chs_conv = dec_chs_conv)
+ #self.head = depthwise_separable_conv(1, 1, padding = "same", kernel_size=1)
+
+
+ def forward(self, x):
+ enc_ftrs = self.encoder(x)
+ out = self.decoder(enc_ftrs[::-1][0], enc_ftrs[::-1][1:])
+ #out = self.head(out)
+ return out
+
+@torch.no_grad()
+def evaluate(model, val_loader):
+ model.eval()
+ outputs = [model.validation_step(batch) for batch in val_loader]
+ return model.validation_epoch_end(outputs)
+
+def fit(epochs, lr, model, train_loader, val_loader, path, opt_func=torch.optim.Adam):
+ history = []
+ optimizer = opt_func(model.parameters(), lr, eps=1e-07)
+ for epoch in range(epochs):
+ # Training Phase
+ model.train()
+ train_losses = []
+ for batch in train_loader:
+ loss = model.training_step(batch)
+ train_losses.append(loss)
+ loss.backward()
+ optimizer.step()
+ optimizer.zero_grad()
+ # Validation phase
+ result = evaluate(model, val_loader)
+ result['train_loss'] = torch.stack(train_losses).mean().item()
+ model.epoch_end(epoch, result)
+ history.append(result)
+ torch.save(model.state_dict(),f'{path}/Unet_dict_V9.pth')
+ torch.save(history,f'{path}/history_V9.pt')
+ return history
+
+def get_default_device():
+ """Pick GPU if available, else CPU"""
+ if torch.cuda.is_available():
+ return torch.device('cuda')
+ else:
+ print('no GPU found')
+ return torch.device('cpu')
+
+def to_device(data, device):
+ """Move tensor(s) to chosen device"""
+ if isinstance(data, (list,tuple)):
+ return [to_device(x, device) for x in data]
+ return data.to(device, non_blocking=True)
+
+class DeviceDataLoader():
+ """Wrap a dataloader to move data to a device"""
+ def __init__(self, dl, device):
+ self.dl = dl
+ self.device = device
+
+ def __iter__(self):
+ """Yield a batch of data after moving it to device"""
+ for b in self.dl:
+ yield to_device(b, self.device)
+
+ def __len__(self):
+ """Number of batches"""
+ return len(self.dl)
+
+def Create_Dataloader(path, batch_size = 100, percent_val = 0.2):
+ dataset = torch.load(path) # create the pytorch dataset
+ #size_data = 500 #shrink dataset for colab
+ #rest = len(dataset) -size_data
+ #dataset,_ = torch.utils.data.random_split(dataset, [size_data, rest])
+ val_size = int(len(dataset) * percent_val)
+ train_size = len(dataset) - val_size
+
+ train_ds, val_ds = random_split(dataset, [train_size, val_size])
+ # Create DataLoader
+ train_dl = DataLoader(train_ds, batch_size, shuffle=True, num_workers=1, pin_memory=True)
+ valid_dl = DataLoader(val_ds, batch_size, num_workers=1, pin_memory=True)
+
+ return train_dl, valid_dl
+
+if __name__ == '__main__':
+ #os.chdir('F:/RWTH/HiWi_IEHK/DAMASK3/UNet/Trainingsdata')
+ use_seeds = False
+ seed = 373686838
+ num_epochs = 1300
+ b_size = 8
+ opt_func = torch.optim.Adam
+ lr = 0.00001
+ kernel = 9
+ print(f'number auf epochs: {num_epochs}')
+ print(f'batchsize: {b_size}')
+ print(f'learning rate: {lr}')
+ print(f'kernel size is: {kernel}')
+ if not use_seeds:
+ seed = random.randrange(2**32 - 1)
+ print(f' seed is: {seed}')
+ torch.manual_seed(seed)
+ random.seed(seed)
+ np.random.seed(seed)
+ device = get_default_device()
+ train_dl, valid_dl = Create_Dataloader('/home/yk138599/Hiwi/damask3/UNet/Trainingsdata/Training_Dataset_normalized_32_V2.pt', batch_size= b_size )
+ train_dl = DeviceDataLoader(train_dl, device)
+ valid_dl = DeviceDataLoader(valid_dl, device)
+
+ model = to_device(UNet(kernel_size=kernel).double(), device)
+ history = fit(num_epochs, lr, model, train_dl, valid_dl,'/home/yk138599/Hiwi/damask3/UNet/output', opt_func)