Update app.py

app.py

@@ -6,28 +6,74 @@ import imageio
 import cv2
 
 
-class Generator(nn.Module):
-    # Refer to the link below for explanations about nc, nz, and ngf
-    # https://pytorch.org/tutorials/beginner/dcgan_faces_tutorial.html#inputs
-    def __init__(self, nc=4, nz=100, ngf=64):
-        super(Generator, self).__init__()
-        self.network = nn.Sequential(
-            nn.ConvTranspose2d(nz, ngf * 4, 3, 1, 0, bias=False),
-            nn.BatchNorm2d(ngf * 4),
-            nn.ReLU(True),
-            nn.ConvTranspose2d(ngf * 4, ngf * 2, 3, 2, 1, bias=False),
-            nn.BatchNorm2d(ngf * 2),
-            nn.ReLU(True),
-            nn.ConvTranspose2d(ngf * 2, ngf, 4, 2, 0, bias=False),
-            nn.BatchNorm2d(ngf),
-            nn.ReLU(True),
-            nn.ConvTranspose2d(ngf, nc, 4, 2, 1, bias=False),
-            nn.Tanh(),
-        )
+class RelationModuleMultiScale(torch.nn.Module):
+
+    def __init__(self, img_feature_dim, num_bottleneck, num_frames):
+        super(RelationModuleMultiScale, self).__init__()
+        self.subsample_num = 3
+        self.img_feature_dim = img_feature_dim
+        self.scales = [i for i in range(num_frames, 1, -1)]
+        self.relations_scales = []
+        self.subsample_scales = []
+        for scale in self.scales:
+            relations_scale = self.return_relationset(num_frames, scale)
+            self.relations_scales.append(relations_scale)
+            self.subsample_scales.append(min(self.subsample_num, len(relations_scale)))
+        self.num_frames = num_frames
+        self.fc_fusion_scales = nn.ModuleList() # high-tech modulelist
+        for i in range(len(self.scales)):
+            scale = self.scales[i]
+            fc_fusion = nn.Sequential(nn.ReLU(), nn.Linear(scale * self.img_feature_dim, num_bottleneck), nn.ReLU())
+            self.fc_fusion_scales += [fc_fusion]
 
     def forward(self, input):
-        output = self.network(input)
-        return output
+        act_scale_1 = input[:, self.relations_scales[0][0] , :]
+        act_scale_1 = act_scale_1.view(act_scale_1.size(0), self.scales[0] * self.img_feature_dim)
+        act_scale_1 = self.fc_fusion_scales[0](act_scale_1)
+        act_scale_1 = act_scale_1.unsqueeze(1)
+        act_all = act_scale_1.clone()
+        for scaleID in range(1, len(self.scales)):
+            act_relation_all = torch.zeros_like(act_scale_1)
+            num_total_relations = len(self.relations_scales[scaleID])
+            num_select_relations = self.subsample_scales[scaleID]
+            idx_relations_evensample = [int(ceil(i * num_total_relations / num_select_relations)) for i in range(num_select_relations)]
+            for idx in idx_relations_evensample:
+                act_relation = input[:, self.relations_scales[scaleID][idx], :]
+                act_relation = act_relation.view(act_relation.size(0), self.scales[scaleID] * self.img_feature_dim)
+                act_relation = self.fc_fusion_scales[scaleID](act_relation)
+                act_relation = act_relation.unsqueeze(1)
+                act_relation_all += act_relation
+            act_all = torch.cat((act_all, act_relation_all), 1)
+        return act_all
+
+    def return_relationset(self, num_frames, num_frames_relation):
+        import itertools
+        return list(itertools.combinations([i for i in range(num_frames)], num_frames_relation))
+
+
+parser = argparse.ArgumentParser()
+parser.add_argument('--dataset',  default='Sprite', help='datasets')
+parser.add_argument('--data_root', default='dataset', help='root directory for data')
+parser.add_argument('--num_class',  type=int, default=15, help='the number of class for jester dataset')
+parser.add_argument('--input_type',  default='image', choices=['feature', 'image'], help='the type of input')
+parser.add_argument('--src',  default='domain_1', help='source domain')
+parser.add_argument('--tar',  default='domain_2', help='target domain')
+parser.add_argument('--num_segments', type=int, default=8, help='the number of frame segment')
+parser.add_argument('--backbone', type=str, default="dcgan", choices=['dcgan', 'resnet101', 'I3Dpretrain','I3Dfinetune'], help='backbone')
+parser.add_argument('--channels', default=3, type=int, help='input channels for image inputs')
+parser.add_argument('--add_fc', default=1, type=int, metavar='M', help='number of additional fc layers (excluding the last fc layer) (e.g. 0, 1, 2)')
+parser.add_argument('--fc_dim', type=int, default=1024, help='dimension of added fc')
+parser.add_argument('--frame_aggregation', type=str, default='trn', choices=[ 'rnn', 'trn'], help='aggregation of frame features (none if baseline_type is not video)')
+parser.add_argument('--dropout_rate', default=0.5, type=float, help='dropout ratio for frame-level feature (default: 0.5)')
+parser.add_argument('--f_dim', type=int, default=512, help='dim of f')
+parser.add_argument('--z_dim', type=int, default=512, help='dimensionality of z_t')
+parser.add_argument('--f_rnn_layers', type=int, default=1, help='number of layers (content lstm)')
+parser.add_argument('--use_bn', type=str, default='none', choices=['none', 'AdaBN', 'AutoDIAL'], help='normalization-based methods')
+parser.add_argument('--prior_sample', type=str, default='random', choices=['random', 'post'], help='how to sample prior')
+parser.add_argument('--batch_size', default=128, type=int, help='-batch size')
+parser.add_argument('--use_attn', type=str, default='TransAttn', choices=['none', 'TransAttn', 'general'], help='attention-mechanism')
+parser.add_argument('--data_threads', type=int, default=5, help='number of data loading threads')
+opt = parser.parse_args(args=[])
 
 
 def display_gif(file_name, save_name):