dataloader problem (MetaPhys_task_Dataset)

dataset/MetaPhysDataset.py

@@ -1,3 +1,5 @@
+import random
+
 import torch
 import numpy as np
 from torchmeta.utils.data import Task, MetaDataset
@@ -25,9 +27,6 @@ def __init__(self, num_shots_tr, num_shots_ts, option='train',
     def __getitem__(self, index):
         if torch.is_tensor(index):
             index = index.tolist()
-
-        self.dataset_transform = ClassSplitter(shuffle=False, num_train_per_class=self.num_shots_tr,
-                                               num_test_per_class=self.num_shots_ts)
         ap = []
         mo = []
         la = []
@@ -39,6 +38,8 @@ def __getitem__(self, index):
 
         task = PersonTask(ap, mo, la)
 
+        self.dataset_transform = ClassSplitter(shuffle=False, num_train_per_class=self.num_shots_tr,
+                                               num_test_per_class=self.num_shots_ts)
         if self.dataset_transform is not None:
             task = self.dataset_transform(task)
 
@@ -47,10 +48,52 @@ def __getitem__(self, index):
     def __len__(self):
         return len(self.label)
 
+class MetaPhys_task_Dataset(MetaDataset):
+    def __init__(self, num_shots_tr, num_shots_ts, option='train',
+                 unsupervised=0,frame_depth=10,
+                 appearance_data=None, motion_data=None, target=None):
+
+        self.transform = transforms.Compose([transforms.ToTensor()])
+        self.num_samples_per_task = num_shots_tr + num_shots_ts
+        self.frame_depth = frame_depth
+        self.option = option
+        self.num_shots_tr = num_shots_tr
+        self.num_shots_ts = num_shots_ts
+        self.unsupervised = unsupervised
+        self.a = appearance_data
+        self.m = motion_data
+        self.label = target
+        self.dataset_transform = ClassSplitter(shuffle=False, num_train_per_class=num_shots_tr,
+                                                   num_test_per_class=num_shots_ts)
+
+    def __getitem__(self, index):
+        if torch.is_tensor(index):
+            index = index.tolist()
+        ap = []
+        mo = []
+        la = []
+
+        for i in range(4):
+            ap.append(self.a[index][i])
+            mo.append(self.m[index][i])
+            la.append(self.label[index][i])
+
+        task = PersonTask(ap, mo, la)
+
+        self.dataset_transform = ClassSplitter(shuffle=False, num_train_per_class=self.num_shots_tr,
+                                               num_test_per_class=self.num_shots_ts)
+        if self.dataset_transform is not None:
+            task = self.dataset_transform(task)
+
+        return task
+
+    def __len__(self):
+        return len(self.label)
 
 class PersonTask(Task):
     def __init__(self, a, m ,label):
         super(PersonTask, self).__init__(None, None) # Regression task
+        self.transform = transforms.Compose([transforms.ToTensor()])
         self.a = a
         self.m = m
         self.label = label

dataset/dataset_loader.py

@@ -4,7 +4,7 @@
 
 from dataset.DeepPhysDataset import DeepPhysDataset
 from dataset.PhysNetDataset import PhysNetDataset
-from dataset.MetaPhysDataset import MetaPhysDataset
+from dataset.MetaPhysDataset import MetaPhysDataset, MetaPhys_task_Dataset
 
 def dataset_loader(save_root_path: str = "/media/hdd1/dy_dataset/",
                    model_name: str = "DeepPhys",
@@ -71,25 +71,32 @@ def dataset_loader(save_root_path: str = "/media/hdd1/dy_dataset/",
                                   )
 
     if model_name == "MetaPhys_task":
-        appearance_data = []
-        motion_data = []
-        target_data = []
-        print(hpy_file.keys())
-        for key in hpy_file.keys(): #subject1, subject10, ...
-            appearance_data.append(hpy_file[key]['preprocessed_video'][:, :, :, -3:])
-            motion_data.append(hpy_file[key]['preprocessed_video'][:, :, :, :3])
-            target_data.append(hpy_file[key]['preprocessed_label'][:])
+        appearance_data_all = []
+        motion_data_all = []
+        target_data_all = []
+
+        for key in hpy_file.keys(): #1, 2, ...
+            appearance_data = []
+            motion_data = []
+            target_data = []
+            for data in hpy_file[key]: #1/1, 1/2, ...
+                appearance_data.append(hpy_file[key][data]['preprocessed_video'][:, :, :, -3:])
+                motion_data.append(hpy_file[key][data]['preprocessed_video'][:, :, :, :3])
+                target_data.append(hpy_file[key][data]['preprocessed_label'][:])
+            appearance_data_all.append(appearance_data) #np.asarray(
+            motion_data_all.append(motion_data)
+            target_data_all.append(target_data)
         hpy_file.close()
 
-        dataset = MetaPhysDataset(num_shots,
+        dataset = MetaPhys_task_Dataset(num_shots,
                                   num_test_shots,
                                   option,
                                   unsupervised,
                                   frame_depth=10,
 
-                                  appearance_data=np.asarray(appearance_data),
-                                  motion_data=np.asarray(motion_data),
-                                  target=np.asarray(target_data)
+                                  appearance_data=np.asarray(appearance_data_all),
+                                  motion_data=np.asarray(motion_data_all),
+                                  target=np.asarray(target_data_all)
                                   )
 
     return dataset

meta_params.json

@@ -50,7 +50,7 @@
     },
     "meta_params":
     {
-        "num_shots" : 6,
+        "num_shots" : 2,
         "num_test_shots" : 2,
         "fs" : 30,
         "fs_comment" : "sampling rate of dataset",

models.py

@@ -20,7 +20,7 @@ def get_model(model_name: str = "DeepPhys"):
         return DeepPhys_DA()
     elif model_name == "PhysNet":
         return PhysNet()
-    elif model_name == "MetaPhys":
+    elif model_name == "MetaPhys" or "MetaPhys_task":
         return TSCAN()
     #elif model_name == "PhysNet_LSTM":
         #return PhysNet_2DCNN_LSTM()
@@ -54,7 +54,7 @@ def summary(model, model_name):
     #elif model_name == "PhysNet" or model_name == "PhysNet_LSTM":
         # torchsummary.summary(model, (3, 32, 128, 128))
        # torchinfo.summary(model,(1, 3, 32, 128, 128))
-    elif model_name == "MetaPhys":
+    elif model_name == "MetaPhys" or "MetaPhys_task":
         print('rrrr')
     else:
         log_warning("use implemented model")

nets/models/MetaPhys.py

@@ -56,7 +56,7 @@ def maml_train(tepoch, model, inner_criterion, outer_criterion, inner_optimizer,
         inputs, targets = batch['train']
         test_inputs, test_targets = batch['test']
 
-        test_losses = 0.0
+        test_losses = []
         optimizer.zero_grad()
         for task_idx, (input, target, test_input, test_target) in enumerate(
                 zip(inputs, targets, test_inputs, test_targets)):
@@ -66,15 +66,16 @@ def maml_train(tepoch, model, inner_criterion, outer_criterion, inner_optimizer,
                     diffopt.step(inner_loss)
                 test_logit = fmodel(test_input)
                 test_loss = outer_criterion(test_logit, test_target)
-                test_losses +=test_loss.item()
+                test_losses.append(test_loss.detach())
                 test_loss.backward()
+
         optimizer.step()
-        losses = test_losses / len(tepoch)
+        losses = sum(test_losses) / len(tepoch)
         tepoch.set_postfix(loss=losses)
 
 def maml_val(tepoch, model, inner_criterion, outer_criterion, inner_optimizer, num_adapt_steps):
     model.train()
-    test_losses = 0.0
+    test_losses = []
     for batch in tepoch:
         tepoch.set_description(f"Validation")
 
@@ -94,8 +95,9 @@ def maml_val(tepoch, model, inner_criterion, outer_criterion, inner_optimizer, n
                     diffopt.step(inner_loss)
                 test_logit = fmodel(test_input).detach()
                 test_loss = outer_criterion(test_logit, test_target)
-                test_losses +=test_loss.item()
-        losses = test_losses / len(tepoch)
+                test_losses.append(test_loss.detach())
+
+        losses = sum(test_losses) / len(tepoch)
         tepoch.set_postfix(loss=losses)
     '''
     if min_val_loss > test_losses:  # save the train model
@@ -113,7 +115,7 @@ def maml_val(tepoch, model, inner_criterion, outer_criterion, inner_optimizer, n
 
 def maml_test(tepoch, model, inner_criterion, outer_criterion, inner_optimizer, num_adapt_steps):
     model.train()
-    test_losses = 0.0
+    mean_test_loss = torch.tensor(0., device='cuda:9')
     inference_array = []
     target_array = []
     for batch in tepoch:
@@ -135,7 +137,7 @@ def maml_test(tepoch, model, inner_criterion, outer_criterion, inner_optimizer,
                     diffopt.step(inner_loss)
                 test_logit = fmodel(test_input).detach()
                 test_loss = outer_criterion(test_logit, test_target)
-                test_losses += test_loss.item()
+                mean_test_loss += test_loss.item()
 
                 inference_array.extend(test_logit.cpu().numpy())
                 target_array.extend(test_target.cpu().numpy())

utils/text_preprocess.py

@@ -1,5 +1,5 @@
 import numpy as np
-
+import h5py
 
 def Deepphys_preprocess_Label(path):
     '''
@@ -45,3 +45,20 @@ def PhysNet_preprocess_Label(path):
     f.close()
 
     return split_raw_label
+
+def cohface_Label(path, frame_total):
+    f = h5py.File(path, "r")
+    label = list(f['pulse'])
+    f.close()
+    label = np.interp(np.arange(0, len(frame_total)+1),
+                      np.linspace(0, len(frame_total)+1, num=len(label)),
+                      label)
+    delta_label = []
+    for i in range(len(label) - 1):
+        delta_label.append(label[i + 1] - label[i])
+    delta_label -= np.mean(delta_label)
+    delta_label /= np.std(delta_label)
+    delta_label = np.array(delta_label).astype('float32')
+    delta_pulse = delta_label.copy()  # 이거 왜 있지?
+
+    return delta_pulse