diff --git a/src/deep_neurographs/machine_learning/graph_models.py b/src/deep_neurographs/machine_learning/graph_models.py
index 095645c..8194e57 100644
--- a/src/deep_neurographs/machine_learning/graph_models.py
+++ b/src/deep_neurographs/machine_learning/graph_models.py
@@ -42,7 +42,8 @@ class MLP(torch.nn.Module):
     def __init__(self, input_channels):
         super().__init__()
         self.linear1 = Linear(input_channels, input_channels // 2)
-        self.linear2 = Linear(input_channels // 2, 1)
+        self.linear2 = Linear(input_channels // 2, input_channels // 2)
+        self.linear3 = Linear(input_channels // 2, 1)
         self.ELU = ELU()
 
     def forward(self, x, edge_index):
@@ -50,4 +51,7 @@ def forward(self, x, edge_index):
         x = self.ELU(x)
         x = F.dropout(x, p=0.25)
         x = self.linear2(x)
+        x = self.ELU(x)
+        x = F.dropout(x, p=0.25)
+        x = self.linear3(x)
         return x
diff --git a/src/deep_neurographs/machine_learning/inference.py b/src/deep_neurographs/machine_learning/inference.py
index 1be9089..7a639a4 100644
--- a/src/deep_neurographs/machine_learning/inference.py
+++ b/src/deep_neurographs/machine_learning/inference.py
@@ -131,7 +131,7 @@ def run_without_seeds(
 
         # Report progress
         if i > progress_cnt * chunk_size and progress_bar:
-            progress_cnt, t1 = report_progress(
+            progress_cnt, t1 = utils.report_progress(
                 i, n_batches, chunk_size, progress_cnt, t0, t1
             )
             t0, t1 = utils.init_timers()
@@ -270,44 +270,41 @@ def run_model(dataset, model, model_type):
             hat_y_i = np.array(hat_y_i)
             hat_y.extend(hat_y_i.tolist())
     else:
+        data = dataset["dataset"]
         hat_y = model.predict_proba(data["inputs"])[:, 1]
     return np.array(hat_y)
 
 
 def run_graph_model(graph_data, model):
     # Run model
+    model.eval()
     x, edge_index = toGPU(graph_data.data)
-    hat_y = model(x, edge_index)
+    with torch.no_grad():
+        hat_y = sigmoid(model(x, edge_index))
 
     # Reformat pred
     idx = graph_data.n_proposals
     hat_y = ml_utils.toCPU(hat_y[0:idx, 0])
-    return ml_utils.sigmoid(hat_y)
-
-
-# Utils
-def report_progress(current, total, chunk_size, cnt, t0, t1):
-    eta = get_eta(current, total, chunk_size, t1)
-    runtime = get_runtime(current, total, chunk_size, t0, t1)
-    utils.progress_bar(current, total, eta=eta, runtime=runtime)
-    return cnt + 1, time()
-
-
-def get_eta(current, total, chunk_size, t0, return_str=True):
-    chunk_runtime = time() - t0
-    remaining = total - current
-    eta = remaining * (chunk_runtime / chunk_size)
-    t, unit = utils.time_writer(eta)
-    return f"{round(t, 4)} {unit}" if return_str else eta
-
+    return hat_y
 
-def get_runtime(current, total, chunk_size, t0, t1):
-    eta = get_eta(current, total, chunk_size, t1, return_str=False)
-    total_runtime = time() - t0 + eta
-    t, unit = utils.time_writer(total_runtime)
-    return f"{round(t, 4)} {unit}"
 
 def toGPU(graph_data):
+    """
+    Moves "graph_data" from CPU to GPU.
+
+    Parameters
+    ----------
+    graph_data : GraphDataset
+        Dataset to be moved to GPU.
+
+    Returns
+    -------
+    x : torch.Tensor
+        Matrix of node feature vectors.
+    edge_idx : torch.Tensor
+        Tensor containing edges in graph.
+
+    """
     x = graph_data.x.to("cuda:0", dtype=torch.float32)
     edge_index = graph_data.edge_index.to("cuda:0")
     return x, edge_index
diff --git a/src/deep_neurographs/reconstruction.py b/src/deep_neurographs/reconstruction.py
index 71d94fd..1bd668b 100644
--- a/src/deep_neurographs/reconstruction.py
+++ b/src/deep_neurographs/reconstruction.py
@@ -31,14 +31,6 @@ def get_accepted_propoals_blocks(
     accepts = dict()
     for block_id in blocks:
         # Get accepts
-        preds = threshold_preds(
-            preds,
-            idx_to_edge,
-            low_threshold,
-            valid_idxs=block_to_idxs[block_id],
-        )
-
-        # Refine accepts wrt structure
         if structure_aware:
             graph = neurographs[block_id].copy()
             accepts[block_id] = get_structure_aware_accepts(
@@ -49,6 +41,12 @@ def get_accepted_propoals_blocks(
                 low_threshold=low_threshold,
             )
         else:
+            preds = threshold_preds(
+                preds,
+                idx_to_edge,
+                low_threshold,
+                valid_idxs=block_to_idxs[block_id],
+            )
             accepts[block_id] = preds.keys()
     return accepts
 
diff --git a/src/deep_neurographs/utils.py b/src/deep_neurographs/utils.py
index e9462ca..5b26bbe 100644
--- a/src/deep_neurographs/utils.py
+++ b/src/deep_neurographs/utils.py
@@ -620,6 +620,32 @@ def time_writer(t, unit="seconds"):
         t, unit = time_writer(t, unit=unit)
     return t, unit
 
+def report_progress(current, total, chunk_size, cnt, t0, t1):
+    eta = get_eta(current, total, chunk_size, t1)
+    runtime = get_runtime(current, total, chunk_size, t0, t1)
+    utils.progress_bar(current, total, eta=eta, runtime=runtime)
+    return cnt + 1, time()
+
+
+def get_eta(current, total, chunk_size, t0, return_str=True):
+    chunk_runtime = time() - t0
+    remaining = total - current
+    eta = remaining * (chunk_runtime / chunk_size)
+    t, unit = utils.time_writer(eta)
+    return f"{round(t, 4)} {unit}" if return_str else eta
+
+
+def get_runtime(current, total, chunk_size, t0, t1):
+    eta = get_eta(current, total, chunk_size, t1, return_str=False)
+    total_runtime = time() - t0 + eta
+    t, unit = utils.time_writer(total_runtime)
+    return f"{round(t, 4)} {unit}"
+
+def toGPU(graph_data):
+    x = graph_data.x.to("cuda:0", dtype=torch.float32)
+    edge_index = graph_data.edge_index.to("cuda:0")
+    return x, edge_index
+
 
 # --- miscellaneous ---
 def get_img_bbox(origin, shape):