From d110ff70dacec4156a3787eb49e7f2448dfb91a5 Mon Sep 17 00:00:00 2001
From: Dayan-Guan <75825029+Dayan-Guan@users.noreply.github.com>
Date: Tue, 28 Sep 2021 15:42:57 +0800
Subject: [PATCH] Add files via upload

---
 davsn/model/accel_deeplabv2.py | 26 +++-----------------------
 1 file changed, 3 insertions(+), 23 deletions(-)

diff --git a/davsn/model/accel_deeplabv2.py b/davsn/model/accel_deeplabv2.py
index a1f2de5..b870619 100644
--- a/davsn/model/accel_deeplabv2.py
+++ b/davsn/model/accel_deeplabv2.py
@@ -42,10 +42,8 @@ def forward(self, x):
             residual = self.downsample(x)
         out += residual
         out = self.relu(out)
-
         return out
 
-
 class ClassifierModule(nn.Module):
     def __init__(self, inplanes, dilation_series, padding_series, num_classes):
         super(ClassifierModule, self).__init__()
@@ -64,7 +62,6 @@ def forward(self, x):
             out += self.conv2d_list[i + 1](x)
         return out
 
-
 class ResNetMulti(nn.Module):
     def __init__(self, block, layers, num_classes, multi_level):
         self.multi_level = multi_level
@@ -116,25 +113,8 @@ def _make_layer(self, block, planes, blocks, stride=1, dilation=1):
         self.inplanes = planes * block.expansion
         for i in range(1, blocks):
             layers.append(block(self.inplanes, planes, dilation=dilation))
-
         return nn.Sequential(*layers)
 
-    # def forward(self, x):
-    #     x = self.conv1(x)
-    #     x = self.bn1(x)
-    #     x = self.relu(x)
-    #     x = self.maxpool(x)
-    #     x = self.layer1(x)
-    #     x = self.layer2(x)
-    #     x = self.layer3(x)
-    #     if self.multi_level:
-    #         x1 = self.layer5(x)  # produce segmap 1
-    #     else:
-    #         x1 = None
-    #     x2 = self.layer4(x)
-    #     x2 = self.layer6(x2)  # produce segmap 2
-    #     return x1, x2
-
     def forward(self, cf, kf, flow, device):
         cf = self.conv1(cf)
         cf = self.bn1(cf)
@@ -175,8 +155,8 @@ def forward(self, cf, kf, flow, device):
         rec_positions = np.zeros(cf.shape)
         for x in range(cf.shape[-1]):
             for y in range(cf.shape[-2]):
-                x_flow = int(round(x + flow_cf[:, 0, y, x][0]))
-                y_flow = int(round(y + flow_cf[:, 1, y, x][0]))
+                x_flow = int(round(x - flow_cf[:, 0, y, x][0]))
+                y_flow = int(round(y - flow_cf[:, 1, y, x][0]))
                 if x_flow >= 0 and x_flow < flow_cf.shape[-1] and y_flow >= 0 and y_flow < flow_cf.shape[-2]:
                     kf_aux_rec[:, :, y_flow, x_flow] = kf_aux_cpu[:, :, y, x]
                     kf_rec[:, :, y_flow, x_flow] = kf_cpu[:, :, y, x]
@@ -185,6 +165,7 @@ def forward(self, cf, kf, flow, device):
         kf_rec = torch.from_numpy(kf_rec)
         rec_positions = torch.from_numpy(rec_positions)
 
+
         pred_aux = self.sf_layer(torch.cat((cf_aux, (rec_positions*kf_aux_rec).float().cuda(device)), dim=1))
         pred = self.sf_layer(torch.cat((cf, (rec_positions*kf_rec).float().cuda(device)), dim=1))
         return pred_aux, pred, cf_aux, cf, kf_aux, kf
@@ -197,7 +178,6 @@ def get_1x_lr_params_no_scale(self):
         any batchnorm parameter
         """
         b = []
-
         b.append(self.conv1)
         b.append(self.bn1)
         b.append(self.layer1)