++

geometry/baseline/generate.py

@@ -10,7 +10,7 @@ def __init__(self):
         self.paths.append("/scratchf/DFC2015/BE_ORTHO_27032011_315130_56865.tif")
 
         tmp = torch.arange(65536).long()
-        self.broad = ((tmp / 256).long(), (tmp % 256).long())
+        self.broad = ((tmp / 256).long().cuda(), (tmp % 256).long().cuda())
 
     def oldCoordinate(self, r, c, a, b, d):
         dr_ = a[0] + r * d[0], a[1] + r * d[1]
@@ -21,17 +21,16 @@ def get(self):
         tirage = torch.rand(6)
 
         i = int(tirage[0] * len(self.paths))
-        image = PIL.Image.open(self.paths[i])
-        image = image.resize((5000, 5000))
-        image = torch.Tensor(numpy.asarray(image))
-
-        r = int(tirage[1] * image.shape[0] - 1030)
-        c = int(tirage[2] * image.shape[1] - 1030)
-        x = image[r : r + 1024, c : c + 1024, :].clone() / 255
+        x = PIL.Image.open(self.paths[i])
+        x = torch.Tensor(numpy.asarray(x)).clone().cuda()
         x = torch.stack([x[:, :, 0], x[:, :, 1], x[:, :, 2]], dim=0)
+        x = torch.nn.functional.avg_pool2d(x, kernel_size=2)
 
-        a = int(tirage[3] * 400 + 300), int(tirage[4] * 400 + 300)
+        r = int(tirage[1] * x.shape[1] - 1030)
+        c = int(tirage[2] * x.shape[2] - 1030)
+        x = x[:, r : r + 1024, c : c + 1024] / 255
 
+        a = int(tirage[3] * 400 + 300), int(tirage[4] * 400 + 300)
         angle = tirage[5] * 3.1415 / 2
 
         dbx = int(256 * torch.cos(angle))
@@ -44,7 +43,7 @@ def get(self):
 
         p = self.oldCoordinate(self.broad[0], self.broad[1], a, b, d)
         p = torch.clamp(p[0], 0, 1023), torch.clamp(p[1], 0, 1023)
-        x_ = torch.zeros(3, 256, 256)
+        x_ = torch.zeros(3, 256, 256).cuda()
         x_[:, self.broad[0], self.broad[1]] = x[:, p[0], p[1]]
 
         return x, x_, (a, b, d)