bug fix in dataset object

MouseProj/dataset.py

@@ -60,8 +60,8 @@ def toTensorBV (sample):
 	# use if you use 370 BV image
 
 	image, label = sample['image'], sample['label']
-	imageTensor = torch.from_numpy(image-0.25) # normalize the image by mean reduction, the mean=0.25 here 
-	labelTensor = torch.from_numpy(label) # no need to conduct anything special to it, I think
+	imageTensor = torch.from_numpy(((image-0.25)*256).copy()) # normalize the image by mean reduction, the mean=0.25 here 
+	labelTensor = torch.from_numpy(label.copy()) # no need to conduct anything special to it, I think
 	labelTensor = torch.round(labelTensor)
 
 	return {'image': imageTensor, 'label': labelTensor}
@@ -97,7 +97,7 @@ def toTensor (sample):
 
 	return {'image': imageTensor, 'label': labelTensor}
 
-def AffineFun(img, xr, yr, zr, xm, ym, zm, s, order):
+def AffineFun(img, xr, yr, zr, xm, ym, zm, s, DS, order):
 	'''
 	Notes:
 		Rotate and move
@@ -122,6 +122,12 @@ def AffineFun(img, xr, yr, zr, xm, ym, zm, s, order):
 	xc = img[0].shape[0]//2
 	yc = img[0].shape[1]//2
 	zc = img[0].shape[2]//2
+
+	xout = np.round(img[0].shape[0]/DS).astype(np.int16)
+	yout = np.round(img[0].shape[1]/DS).astype(np.int16)
+	zout = np.round(img[0].shape[2]/DS).astype(np.int16)
+
+	imgout = np.zeros((xout, yout, zout))
 
 	Mc = np.array([ [1, 0, 0, xc],
 					[0, 1, 0, yc],
@@ -132,7 +138,12 @@ def AffineFun(img, xr, yr, zr, xm, ym, zm, s, order):
 					[0, 1/s, 0, 0],
 					[0, 0, 1/s, 0], 
 					[0, 0, 0, 1]])
-
+
+	MD = np.array([ [DS, 0, 0, 0],
+					[0, DS, 0, 0],
+					[0, 0, DS, 0], 
+					[0, 0, 0, 1]])
+
 	Rx = np.array([ [cosx, sinx, 0, 0],
 					[-sinx, cosx, 0, 0],
 					[0, 0, 1, 0], 
@@ -158,9 +169,9 @@ def AffineFun(img, xr, yr, zr, xm, ym, zm, s, order):
 					[0, 0, 1, zm],
 					[0 ,0, 0, 1]])
 
-	Matrix = np.linalg.multi_dot([Mc, Ms, Rx, Ry, Rz, Mb, MM])
-	img[0] = affine_transform(img[0], Matrix, output_shape=img[0].shape, order=order)
-	return img
+	Matrix = np.linalg.multi_dot([Mc, Ms, Rx, Ry, Rz, Mb, MM, MD])
+	imgout = affine_transform(img[0], Matrix, output_shape=(xout, yout, zout), order=order)
+	return imgout[None,:,:,:]
 
 def filpFun(img, x, y, z):
 	'''
@@ -248,11 +259,12 @@ class RandomAffine(object):
 	'''
 	Random rotation and move
 	'''
-	def __init__(self, fluR=0, fluM=0, fluS=1):
+	def __init__(self, fluR=0, fluM=0, fluS=1, DS=1):
 
 		self.fluR = fluR
 		self.fluM = fluM
 		self.fluS = fluS
+		self.DS = DS
 
 	def __call__(self, sample):
 
@@ -261,8 +273,8 @@ def __call__(self, sample):
 		s = np.random.uniform(1/self.fluS, self.fluS, size=1)
 
 		image, label = sample['image'], sample['label']
-		return {'image': AffineFun(image, xr, yr, zr, xm, ym, zm, s, 3), \
-				'label': AffineFun(label, xr, yr, zr, xm, ym, zm, s, 0)}
+		return {'image': AffineFun(image, xr, yr, zr, xm, ym, zm, s, self.DS, 3), \
+				'label': AffineFun(label, xr, yr, zr, xm, ym, zm, s, self.DS, 0)}
 
 class niiDataset(Dataset):
 	'''
@@ -404,7 +416,7 @@ def __getitem__(self, indice):
 		BBox = torch.from_numpy(BBox)
 		# Get the BBox ground truth
 
-		sample = toTensor(sample)
+		sample = toTensorBV(sample)
 		imageTensor = sample['image']
 
 		sample = {'image':imageTensor, 'label':BBox}

MouseProj/vnet.py

@@ -209,11 +209,11 @@ def __init__(self, img_size, out_size=6, elu=True):
 		x, y, z = img_size
 
 		self.in_tr = InputTransition(16, elu)
-		self.down_tr32 = DownTransition(16, 2, elu, dropout=False) # /2
-		self.down_tr64 = DownTransition(32, 3, elu, dropout=False) # /4
+		self.down_tr32 = DownTransition(16, 2, elu, dropout=True) # /2
+		self.down_tr64 = DownTransition(32, 3, elu, dropout=True) # /4
 		self.down_tr128 = DownTransition(64, 3, elu, dropout=True) # /8
 		self.down_tr256 = DownTransition(128, 3, elu, dropout=True) # /16
-		self.down_tr512 = DownTransition(256, 4, elu, dropout=True, res=True) # /32, 8 res fcn layers
+		self.down_tr512 = DownTransition(256, 6, elu, dropout=True, res=True) # /32, 8 res fcn layers
 
 		self.gap = nn.AvgPool3d(kernel_size = (x//32,y//32,z//32)) # N, C, 1, 1, 1
 
@@ -238,7 +238,7 @@ def forward(self, x):
 
 		out = self.fc1(out)
 		out = self.fc2(out)
-
+        
 		return out
 
 

Untitled1.ipynb

@@ -0,0 +1,101 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": 1,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import torch as tf"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 2,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "'GeForce GTX 1080'"
+      ]
+     },
+     "execution_count": 2,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "tf.cuda.get_device_name(0)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 3,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import time"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "%%capture output \n",
+    "while True:\n",
+    "    print (time.time())"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 6,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "'GeForce GTX 1080'"
+      ]
+     },
+     "execution_count": 6,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "tf.cuda.get_device_name(0)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": []
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.7.3"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}