fix fetal fault in eval

data/dataset.py

@@ -89,7 +89,7 @@ class Dataset():
         #TODO: check whose stride is negative to fix this instead copy all 
         # some of the strides of a given numpy array are negative.
         # This is currently not supported, but will be added in future releases.
-        return img.copy(), bbox.copy(), label.copy(), scale,ori_img
+        return img.copy(), bbox.copy(), label.copy(), scale, ori_img
 
     def __len__(self):
         return len(self.db)

model/faster_rcnn.py

@@ -370,13 +370,13 @@ class FasterRCNN(nn.Module):
         self.lr3=self.optimizer.param_groups[2]['lr'] = lr3
         return self.optimizer
 
-    def get_optimizer(self):
+    def get_optimizer_(self):
         self.lr1=lr = opt.lr1
         params = [param for param in self.parameters() if param.requires_grad]
         self.optimizer = t.optim.SGD(params, lr=lr,momentum=0.9,weight_decay=0.0005)
         return self.optimizer
 
-    def get_optimizer_(self):
+    def get_optimizer(self):
         lr = opt.lr1
         self.lr1= lr
         params = []
@@ -385,7 +385,7 @@ class FasterRCNN(nn.Module):
                 if 'bias' in key:
                     params += [{'params':[value],'lr':lr*2, 'weight_decay':  0}]
                 else:
-                    params += [{'params':[value],'lr':lr, 'weight_decay': 5e-4}]
+                    params += [{'params':[value],'lr':lr, 'weight_decay':opt.weight_decay}]
         self.optimizer = t.optim.SGD(params, momentum=0.9)
         return self.optimizer
 

model/faster_rcnn_vgg16.py

@@ -13,21 +13,15 @@ def decom_vgg16(pretrained=True):
     model = vgg16(pretrained)
     features = list(model.features)[:30]
     classifier = model.classifier
-    classifier = list(classifier)
-    # delete dropout
-    del classifier[6]
-    del classifier[5]
-    del classifier[2]
-    classifier = nn.Sequential(*classifier)
-    # 
-    # del classifier._modules['6']
+    # classifier = list(classifier)
+    # del the last layer
+    del classifier._modules['6']
 
-    # 冻结前几层的卷积
+    # free top3 conv
     for layer in features[:10]:
         for p in layer.parameters():
             p.requires_grad=False
 
-        
     return nn.Sequential(*features),classifier
 
 def decom_vgg16_chainer(pretrained=True):
@@ -41,6 +35,10 @@ def decom_vgg16_chainer(pretrained=True):
     del classifier[5]
     del classifier[2]
     classifier = nn.Sequential(*classifier)
+
+    # chainer ceil mode = True
+    for idx in [4,9,16,23]:
+        features[idx].ceil_mode=True
     # 
     # del classifier._modules['6']
 
@@ -58,6 +56,8 @@ def decom_vgg16bn(pretrained=True):
     classifier = model.classifier
     del classifier._modules['6']
 
+
+
     # 冻结前几层的卷积
     for layer in features[:13]:
         for p in layer.parameters():
@@ -130,7 +130,7 @@ class FasterRCNNVGG16(FasterRCNN):
                  min_size=600, max_size=1000,
                  ratios=[0.5, 1, 2], anchor_scales=[8, 16, 32]
                  ):
-        extractor,classifier = decom_vgg16_chainer(not opt.load_path)
+        extractor,classifier = decom_vgg16(not opt.load_path)
 
         rpn = RegionProposalNetwork(
             512, 512,

train.py

@@ -33,7 +33,7 @@ def eval(dataloader,faster_rcnn,test_num=1000):
         if ii==test_num:break
 
     result = eval_detection_voc(
-        pred_bboxes, pred_labels, pred_labels,
+        pred_bboxes, pred_labels, pred_scores,
         gt_bboxes, gt_labels, gt_difficults,
         use_07_metric=True)
     return result
@@ -70,7 +70,7 @@ def train(**kwargs):
             scale = at.scalar(scale)
             img,bbox,label = img.cuda().float(),bbox_.cuda(),label_.cuda()
             img,bbox,label = Variable(img),Variable(bbox),Variable(label)
-            losses,rois = trainer.train_step(img,bbox,label,scale)
+            losses = trainer.train_step(img,bbox,label,scale)
             
             if (ii+1)%opt.plot_every == 0:
                 if os.path.exists(opt.debug_file):

trainer.py

@@ -140,11 +140,11 @@ class FasterRCNNTrainer(nn.Module):
         # I think it's fine to break the computation graph of rois
         # Sample RoIs and forward
         sample_roi, gt_roi_loc, gt_roi_label = self.proposal_target_creator(
-                                roi, 
-                                at.tonumpy(bbox),
-                                at.tonumpy(label),
-                                self.loc_normalize_mean, 
-                                self.loc_normalize_std)
+                                    roi, 
+                                    at.tonumpy(bbox),
+                                    at.tonumpy(label),
+                                    self.loc_normalize_mean, 
+                                    self.loc_normalize_std)
         self.sample_roi, self.gt_roi_label = sample_roi,gt_roi_label
         #NOTE it's all zero because now it only support for batch=1 now
         sample_roi_index = t.zeros(len(sample_roi))
@@ -199,16 +199,16 @@ class FasterRCNNTrainer(nn.Module):
         losses = [rpn_loc_loss , rpn_cls_loss , roi_loc_loss , roi_cls_loss]
         losses = losses + [sum(losses)]
 
-        return LossTuple(*losses),rois
+        return LossTuple(*losses)
 
     def train_step(self, imgs, bboxes, labels, scale):
         self.optimizer.zero_grad()
-        losses,rois = self.forward(imgs, bboxes, labels, scale)
+        losses = self.forward(imgs, bboxes, labels, scale)
         rpn_loc_loss, rpn_cls_loss, roi_loc_loss, roi_cls_loss,total_loss = losses
         total_loss.backward()
         self.optimizer.step()
         self.update_meters(losses)
-        return losses,rois
+        return losses
 
     # def visulize(self):