model.py

@@ -71,16 +71,18 @@ class FastRCNN(nn.Module):
         # hidden_dim -> hidden_dim.                                                  #
         ##############################################################################
         # Replace "pass" statement with your code
-        self.shared_fc = nn.Sequential(
+        self.cls_head = nn.Sequential(
             nn.Linear(in_dim, hidden_dim),
             nn.Dropout(drop_ratio),
             nn.ReLU(),
-            nn.Linear(hidden_dim, hidden_dim)
+            nn.Linear(hidden_dim, num_classes+1)
+        )
+        self.bbox_head = nn.Sequential(
+            nn.Linear(in_dim, hidden_dim),
+            nn.Dropout(drop_ratio),
+            nn.ReLU(),
+            nn.Linear(hidden_dim, 4)
         )
-
-        self.cls_head = nn.Linear(hidden_dim, self.num_classes+1) # The cls head is a Linear layer that predicts num_classes + 1 (background).
-
-        self.bbox_head = nn.Linear(hidden_dim, 4)# The det head is a Linear layer that predicts offsets(dim=4).
         ##############################################################################
         #                               END OF YOUR CODE                             #
         ##############################################################################
@@ -137,9 +139,8 @@ class FastRCNN(nn.Module):
         # print(feat.shape)
 
         # forward heads, get predicted cls scores & offsets
-        shared_feat = self.shared_fc(feat)
-        cls_scores=self.cls_head(shared_feat)
-        bbox_offsets=self.bbox_head(shared_feat)
+        cls_scores=self.cls_head(feat)
+        bbox_offsets=self.bbox_head(feat)
         # print(cls_scores.shape, bbox_offsets.shape)
 
         # assign targets with proposals
@@ -215,11 +216,11 @@ class FastRCNN(nn.Module):
         # perform RoI Pool & mean pool
         feat=torchvision.ops.roi_pool(feat, torch.cat((proposal_batch_ids.unsqueeze(1), proposals),dim=1), output_size=(self.roi_output_w, self.roi_output_h))
         feat = feat.mean(dim=[2, 3])
-        shared_feat = self.shared_fc(feat)
+
         # forward heads, get predicted cls scores & offsets
-        cls_scores = self.cls_head(shared_feat)
+        cls_scores = self.cls_head(feat)
         # print(cls_scores.shape)
-        bbox_offsets = self.bbox_head(shared_feat)
+        bbox_offsets = self.bbox_head(feat)
         # print(bbox_offsets.shape)
         # get predicted boxes & class label & confidence probability
         proposals = generate_proposal(proposals, bbox_offsets)

utils.py

@@ -234,27 +234,9 @@ def generate_proposal(anchors, offsets):
     # compute the proposal coordinates using the transformation formulas above.  #
     ##############################################################################
     # Replace "pass" statement with your code
-    x1, y1, x2, y2 =anchors[:, 0], anchors[:, 1], anchors[:, 2], anchors[:, 3] 
-  
-    pw = x2 - x1
-    ph = y2 - y1
-    
-    px = x1 + 0.5 * pw
-    py = y1 + 0.5 * ph
-    
-    tx, ty, tw, th = offsets[:, 0], offsets[:, 1], offsets[:, 2], offsets[:, 3]
-    
-    proposal_x = px + tx * pw
-    proposal_y = py + ty * ph
-    proposal_w = pw * torch.exp(tw)
-    proposal_h = ph * torch.exp(th)
-    
-    proposal_x1 = proposal_x - 0.5 * proposal_w
-    proposal_y1 = proposal_y - 0.5 * proposal_h
-    proposal_x2 = proposal_x + 0.5 * proposal_w
-    proposal_y2 = proposal_y + 0.5 * proposal_h
-    
-    proposals = torch.stack((proposal_x1, proposal_y1, proposal_x2, proposal_y2),dim=1)
+    proposals = torch.zeros_like(anchors)
+    proposals[:, :2] = anchors[:, :2] + offsets[:, :2] * (anchors[:, 2:4] - anchors[:, :2])
+    proposals[:, 2:4] = anchors[:, 2:4] * torch.exp(offsets[:, 2:4])
 
 
     ##############################################################################