fine tune on alexnet

panthersuper · Dec 1, 2017 · b728266 · b728266
1 parent c755fb6
commit b728266
Show file tree

Hide file tree

Showing 7 changed files with 170 additions and 28 deletions.
diff --git a/.gitignore b/.gitignore
@@ -1 +1,2 @@
-data/*
+data/*
+recon0/*
diff --git a/main.py b/main.py
@@ -10,14 +10,14 @@
 
 
 # Dataset Parameters
-load_size =128
-fine_size = 113
+load_size =256
+fine_size = 224
 data_mean = np.asarray([0,0,0,0])
-batch_size = 3
+batch_size = 8
 voxel_size = 256
 
 # Training Parameters
-learning_rate = 0.01
+learning_rate = 0.1
 training_epoches = 10
 step_display = 100
 step_save = 2
@@ -113,6 +113,7 @@ def weights_init(m):
         # inputs, labels = Variable(inputs), Variable(labels)
         inputs, labels= Variable(inputs.cuda()), Variable(labels.cuda())
 
+
         # zero the parameter gradients
         optimizer.zero_grad()
 

diff --git a/model/__pycache__/alexnet.cpython-36.pyc b/model/__pycache__/alexnet.cpython-36.pyc
diff --git a/model/__pycache__/reconNet.cpython-36.pyc b/model/__pycache__/reconNet.cpython-36.pyc
diff --git a/model/alexnet.py b/model/alexnet.py
@@ -0,0 +1,58 @@
+import torch.nn as nn
+import torch.utils.model_zoo as model_zoo
+
+
+__all__ = ['AlexNet', 'alexnet']
+
+
+model_urls = {
+    'alexnet': 'https://download.pytorch.org/models/alexnet-owt-4df8aa71.pth',
+}
+
+
+class AlexNet(nn.Module):
+
+    def __init__(self, num_classes=1000):
+        super(AlexNet, self).__init__()
+        self.features = nn.Sequential(
+            nn.Conv2d(3, 64, kernel_size=11, stride=4, padding=2),
+            nn.ReLU(inplace=True),
+            nn.MaxPool2d(kernel_size=3, stride=2),
+            nn.Conv2d(64, 192, kernel_size=5, padding=2),
+            nn.ReLU(inplace=True),
+            nn.MaxPool2d(kernel_size=3, stride=2),
+            nn.Conv2d(192, 384, kernel_size=3, padding=1),
+            nn.ReLU(inplace=True),
+            nn.Conv2d(384, 256, kernel_size=3, padding=1),
+            nn.ReLU(inplace=True),
+            nn.Conv2d(256, 256, kernel_size=3, padding=1),
+            nn.ReLU(inplace=True),
+            nn.MaxPool2d(kernel_size=3, stride=2),
+        )
+        self.classifier = nn.Sequential(
+            nn.Dropout(),
+            nn.Linear(256 * 6 * 6, 4096),
+            nn.ReLU(inplace=True),
+            nn.Dropout(),
+            nn.Linear(4096, 4096),
+            nn.ReLU(inplace=True),
+            nn.Linear(4096, num_classes),
+        )
+
+    def forward(self, x):
+        x = self.features(x)
+        x = x.view(x.size(0), 256 * 6 * 6)
+        # x = self.classifier(x)
+        return x
+
+
+def alex_net(pretrained=False, **kwargs):
+    r"""AlexNet model architecture from the
+    `"One weird trick..." <https://arxiv.org/abs/1404.5997>`_ paper.
+    Args:
+        pretrained (bool): If True, returns a model pre-trained on ImageNet
+    """
+    model = AlexNet(**kwargs)
+    if pretrained:
+        model.load_state_dict(model_zoo.load_url(model_urls['alexnet']))
+    return model
diff --git a/model/reconNet.py b/model/reconNet.py
@@ -1,31 +1,15 @@
 import torch.nn as nn
 
+import sys, os
+sys.path.append('model/')
+
+from alexnet import alex_net
+
 class ReconNet(nn.Module):
 
     def __init__(self):
         super(ReconNet, self).__init__()
-        self.features = nn.Sequential( # imput 113*113 image
-            # encoding process
-            nn.Conv2d(3, 64, kernel_size=5, stride=2, padding=2),
-            nn.BatchNorm2d(64),
-            nn.ReLU(inplace=True),
-            nn.MaxPool2d(kernel_size=3, stride=2),
-            nn.Conv2d(64, 192, kernel_size=5, padding=2),
-            nn.BatchNorm2d(192),
-            nn.ReLU(inplace=True),
-            nn.MaxPool2d(kernel_size=3, stride=2),
-            nn.Conv2d(192, 384, kernel_size=3, padding=1),
-            nn.BatchNorm2d(384),
-            nn.ReLU(inplace=True),
-            nn.Conv2d(384, 256, kernel_size=3, padding=1),
-            nn.BatchNorm2d(256),
-            nn.ReLU(inplace=True),
-            nn.Conv2d(256, 256, kernel_size=3, padding=1),
-            nn.BatchNorm2d(256),
-            nn.ReLU(inplace=True),
-            nn.MaxPool2d(kernel_size=3, stride=2),
-        )
-
+        self.features = alex_net(pretrained=True)
 
         self.latentV = nn.Sequential(
             nn.Dropout(),
@@ -66,7 +50,7 @@ def __init__(self):
 
     def forward(self, x):
         x = self.features(x)
-        x = x.view(x.size(0), 256 * 6 * 6)
+        # x = x.view(x.size(0), 256 * 6 * 6)
         x = self.latentV(x)  # latent vector, size:4096
         x = x.view(x.size(0),512,2,2,2) # reshape to 2 by 2 by 2 cube with 512 channels
         x = self.decoding(x) # convert to 3D voxel distribution

diff --git a/recon0/log.txt b/recon0/log.txt
@@ -0,0 +1,98 @@
+PLACES TRAINING Epoch: 1 100 loss: 0.054
+PLACES TRAINING Epoch: 1 200 loss: 0.049
+PLACES TRAINING Epoch: 1 300 loss: 0.054
+PLACES TRAINING Epoch: 1 400 loss: 0.052
+PLACES TRAINING Epoch: 1 500 loss: 0.057
+PLACES TRAINING Epoch: 1 600 loss: 0.052
+PLACES TRAINING Epoch: 1 700 loss: 0.048
+PLACES TRAINING Epoch: 1 800 loss: 0.049
+PLACES TRAINING Epoch: 1 900 loss: 0.056
+PLACES TRAINING Epoch: 1 1000 loss: 0.053
+PLACES TRAINING Epoch: 1 1100 loss: 0.056
+PLACES TRAINING Epoch: 1 1200 loss: 0.052
+PLACES TRAINING Epoch: 1 1300 loss: 0.051
+PLACES TRAINING Epoch: 1 1400 loss: 0.057
+PLACES TRAINING Epoch: 1 1500 loss: 0.052
+PLACES TRAINING Epoch: 1 1600 loss: 0.056
+PLACES TRAINING Epoch: 1 1700 loss: 0.053
+PLACES TRAINING Epoch: 1 1800 loss: 0.054
+PLACES TRAINING Epoch: 1 1900 loss: 0.056
+PLACES TRAINING Epoch: 1 2000 loss: 0.062
+PLACES TRAINING Epoch: 1 2100 loss: 0.055
+PLACES TRAINING Epoch: 1 2200 loss: 0.056
+PLACES TRAINING Epoch: 1 2300 loss: 0.051
+PLACES TRAINING Epoch: 1 2400 loss: 0.051
+PLACES TRAINING Epoch: 1 2500 loss: 0.055
+PLACES TRAINING Epoch: 1 2600 loss: 0.056
+PLACES TRAINING Epoch: 1 2700 loss: 0.052
+PLACES TRAINING Epoch: 1 2800 loss: 0.050
+PLACES TRAINING Epoch: 1 2900 loss: 0.051
+PLACES TRAINING Epoch: 1 3000 loss: 0.053
+PLACES TRAINING Epoch: 1 3100 loss: 0.056
+PLACES TRAINING Epoch: 1 3200 loss: 0.056
+PLACES TRAINING Epoch: 1 3300 loss: 0.058
+PLACES TRAINING Epoch: 1 3400 loss: 0.055
+PLACES TRAINING Epoch: 1 3500 loss: 0.055
+PLACES TRAINING Epoch: 1 3600 loss: 0.049
+PLACES TRAINING Epoch: 1 3700 loss: 0.047
+PLACES TRAINING Epoch: 1 3800 loss: 0.054
+PLACES TRAINING Epoch: 1 3900 loss: 0.061
+PLACES TRAINING Epoch: 1 4000 loss: 0.054
+PLACES TRAINING Epoch: 2 100 loss: 0.053
+PLACES TRAINING Epoch: 2 200 loss: 0.049
+PLACES TRAINING Epoch: 2 300 loss: 0.054
+PLACES TRAINING Epoch: 2 400 loss: 0.048
+PLACES TRAINING Epoch: 2 500 loss: 0.060
+PLACES TRAINING Epoch: 2 600 loss: 0.052
+PLACES TRAINING Epoch: 2 700 loss: 0.054
+PLACES TRAINING Epoch: 2 800 loss: 0.055
+PLACES TRAINING Epoch: 2 900 loss: 0.052
+PLACES TRAINING Epoch: 2 1000 loss: 0.053
+PLACES TRAINING Epoch: 2 1100 loss: 0.056
+PLACES TRAINING Epoch: 2 1200 loss: 0.057
+PLACES TRAINING Epoch: 2 1300 loss: 0.049
+PLACES TRAINING Epoch: 2 1400 loss: 0.049
+PLACES TRAINING Epoch: 2 1500 loss: 0.050
+PLACES TRAINING Epoch: 2 1600 loss: 0.053
+PLACES TRAINING Epoch: 2 1700 loss: 0.056
+PLACES TRAINING Epoch: 2 1800 loss: 0.049
+PLACES TRAINING Epoch: 2 1900 loss: 0.054
+PLACES TRAINING Epoch: 2 2000 loss: 0.054
+PLACES TRAINING Epoch: 2 2100 loss: 0.064
+PLACES TRAINING Epoch: 2 2200 loss: 0.064
+PLACES TRAINING Epoch: 2 2300 loss: 0.060
+PLACES TRAINING Epoch: 2 2400 loss: 0.058
+PLACES TRAINING Epoch: 2 2500 loss: 0.057
+PLACES TRAINING Epoch: 2 2600 loss: 0.060
+PLACES TRAINING Epoch: 2 2700 loss: 0.053
+PLACES TRAINING Epoch: 2 2800 loss: 0.062
+PLACES TRAINING Epoch: 2 2900 loss: 0.048
+PLACES TRAINING Epoch: 2 3000 loss: 0.055
+PLACES TRAINING Epoch: 2 3100 loss: 0.057
+PLACES TRAINING Epoch: 2 3200 loss: 0.051
+PLACES TRAINING Epoch: 2 3300 loss: 0.063
+PLACES TRAINING Epoch: 2 3400 loss: 0.060
+PLACES TRAINING Epoch: 2 3500 loss: 0.049
+PLACES TRAINING Epoch: 2 3600 loss: 0.053
+PLACES TRAINING Epoch: 2 3700 loss: 0.054
+PLACES TRAINING Epoch: 2 3800 loss: 0.049
+PLACES TRAINING Epoch: 2 3900 loss: 0.051
+PLACES TRAINING Epoch: 2 4000 loss: 0.054
+PLACES TRAINING Epoch: 3 100 loss: 0.055
+PLACES TRAINING Epoch: 3 200 loss: 0.054
+PLACES TRAINING Epoch: 3 300 loss: 0.054
+PLACES TRAINING Epoch: 3 400 loss: 0.054
+PLACES TRAINING Epoch: 3 500 loss: 0.049
+PLACES TRAINING Epoch: 3 600 loss: 0.052
+PLACES TRAINING Epoch: 3 700 loss: 0.058
+PLACES TRAINING Epoch: 3 800 loss: 0.052
+PLACES TRAINING Epoch: 3 900 loss: 0.059
+PLACES TRAINING Epoch: 3 1000 loss: 0.055
+PLACES TRAINING Epoch: 3 1100 loss: 0.056
+PLACES TRAINING Epoch: 3 1200 loss: 0.054
+PLACES TRAINING Epoch: 3 1300 loss: 0.051
+PLACES TRAINING Epoch: 3 1400 loss: 0.055
+PLACES TRAINING Epoch: 3 1500 loss: 0.058
+PLACES TRAINING Epoch: 3 1600 loss: 0.060
+PLACES TRAINING Epoch: 3 1700 loss: 0.050
+PLACES TRAINING Epoch: 3 1800 loss: 0.050