@@ -0,0 +153,4 @@
+                n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
+                m.weight.data.normal_adapter(0, math.sqrt(2. / n))
+            elif isinstance(m, BatchNorm2d):
+                m.weight.data.fill_(1)

@@ -0,0 +96,4 @@
+        self.conv3 = nn.Conv2d(planes, planes * 4, kernel_size=1, bias=False)
+        self.bn3 = BatchNorm2d(planes * 4)
+        self.relu = nn.ReLU(inplace=False)   # for graph mode accelerating
+        # self.relu = nn.ReLU(inplace=True)

@@ -0,0 +51,4 @@
+            return loss_dict['loss'], loss_dict['loss_shrink_maps'], loss_dict['loss_threshold_maps'], loss_dict['loss_binary_maps'], preds
+        
+        # for graph mode accelerating
+        # def forward_fn(bimg, bsma, bsmas, btma, btms):

@@ -0,0 +54,4 @@
+        # def forward_fn(bimg, bsma, bsmas, btma, btms):
+        #     preds = model(bimg)
+        #     m_l, m_lsm, m_ltm, m_lbm = self.criterion(preds, bsma, bsmas, btma, btms)
+        #     return m_l, m_lsm, m_ltm, m_lbm

@@ -0,0 +32,4 @@
+
+        loss_shrink_maps = self.bce_loss(shrink_maps, batch['shrink_map'], batch['shrink_mask'])
+        loss_threshold_maps = self.l1_loss(threshold_maps, batch['threshold_map'], batch['threshold_mask'])
+        metrics = dict(loss_shrink_maps=loss_shrink_maps, loss_threshold_maps=loss_threshold_maps)

@@ -0,0 +48,4 @@
+            self.with_cuda = True
+            # torch.backends.cudnn.benchmark = True
+            self.device = torch.device("cuda")
+            # torch.cuda.manual_seed(self.config['trainer']['seed'])  # 为当前GPU设置随机种子

@@ -0,0 +63,4 @@
+            self.optimizer(grads)
+            return loss, loss_shrink_maps, loss_threshold_maps, loss_binary_maps, preds
+        
+        # for graph mode accelerating