tiny modification for optimizer

config.py

@@ -18,7 +18,7 @@ class Config:
     roi_sigma = 1.
 
     # param for optimizer
-    weight_decay = 0.0001  # NOTE:it's modified
+    weight_decay = 0.0005  # 0.0005 in origin paper but 0.0001 in tf-faster-rcnn
     lr_decay = 0.1  # 1e-3 -> 1e-4
     # lr = 1e-3
     lr1 = 1e-3  # extractor
@@ -40,7 +40,7 @@ class Config:
     # change lr
     milestone = [0, 1, 5, 10]
 
-
+    use_adam = False
 
     # debug
     debug_file = '/tmp/debugf'

data/dataset.py

@@ -82,7 +82,6 @@ class Dataset():
         img, bbox, label, scale = self.tsf((ori_img, bbox, label))
         # 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
 
     def __len__(self):
@@ -90,34 +89,15 @@ class Dataset():
 
 
 class TestDataset():
-    def __init__(self, opt):
+    def __init__(self, opt,split='test',use_difficult=True):
         self.opt = opt
-        self.db = testset = VOCBboxDataset(opt.voc_data_dir, split='test', use_difficult=True)
+        self.db = testset = VOCBboxDataset(opt.voc_data_dir, split=split, use_difficult=use_difficult)
 
     def __getitem__(self, idx):
         ori_img, bbox, label, difficult = self.db.get_example(idx)
         img = preprocess(ori_img)
         return (img), ori_img.shape[1:], bbox, label, difficult
-        # 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.
 
     def __len__(self):
         return len(self.db)
 
-
-class TestDataset2():
-    def __init__(self, opt):
-        self.opt = opt
-        self.db = testset = VOCBboxDataset(opt.voc_data_dir, split='trainval', use_difficult=True)
-
-    def __getitem__(self, idx):
-        ori_img, bbox, label, difficult = self.db.get_example(idx)
-        img = preprocess(ori_img)
-        return (img), ori_img.shape[1:], bbox, label, difficult
-        # 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.
-
-    def __len__(self):
-        return len(self.db)

model/faster_rcnn.py

@@ -66,16 +66,14 @@ class FasterRCNN(nn.Module):
     Region Proposal Networks. NIPS 2015.
 
     Args:
-        extractor (callable Chain): A callable that takes a BCHW image
+        extractor (nn.Module): A module that takes a BCHW image
             array and returns feature maps.
-        rpn (callable Chain): A callable that has the same interface as
+        rpn (nn.Module): A module that has the same interface as
             :class:`~chainercv.links.model.faster_rcnn.RegionProposalNetwork`.
             Please refer to the documentation found there.
-        head (callable Chain): A callable that takes
+        head (nn.Module): A callable that takes
             a BCHW array, RoIs and batch indices for RoIs. This returns class
             dependent localization paramters and class scores.
-        mean (numpy.ndarray): A value to be subtracted from an image
-            in :meth:`prepare`.
         loc_normalize_mean (tuple of four floats): Mean values of
             localization estimates.
         loc_normalize_std (tupler of four floats): Standard deviation
@@ -399,3 +397,16 @@ class FasterRCNN(nn.Module):
         for param_group in self.optimizer.param_groups:
             param_group['lr'] *= decay
         return self.optimizer
+
+    def get_optimizer_adam(self):
+        lr = opt.lr1 *0.1
+        self.lr1 = lr
+        params = []
+        for key, value in dict(self.named_parameters()).items():
+            if value.requires_grad:
+                if 'bias' in key:
+                    params += [{'params': [value], 'lr': lr * 2, 'weight_decay': 0}]
+                else:
+                    params += [{'params': [value], 'lr': lr, 'weight_decay': opt.weight_decay}]
+        self.optimizer = t.optim.Adam(params)
+        return self.optimizer

train.py

@@ -99,7 +99,9 @@ def train(**kwargs):
                 trainer.vis.text(str(trainer.rpn_cm.value().tolist()), win='rpn_cm')
                 # roi confusion matrix
                 trainer.vis.img('roi_cm', at.totensor(trainer.roi_cm.conf, False).float())
-
+        if best_map>0.6: 
+            opt.test_num=10000
+            best_map = 0
         eval_result = eval(test_dataloader, faster_rcnn, test_num=opt.test_num)
 
         if eval_result['map'] > best_map:

trainer.py

@@ -55,6 +55,8 @@ class FasterRCNNTrainer(nn.Module):
         self.loc_normalize_std = faster_rcnn.loc_normalize_std
 
         self.optimizer = self.faster_rcnn.get_optimizer()
+        if opt.use_adam:
+            self.optimizer = self.faster_rcnn.get_optimizer_adam()
 
         # visdom wrapper
         self.vis = Visualizer(env=opt.env)
@@ -198,7 +200,7 @@ class FasterRCNNTrainer(nn.Module):
             save_dict['optimizer'] = self.optimizer.state_dict()
 
         if save_path is None:
-            timestr = time.strftime('%m%d-%H%M')
+            timestr = time.strftime('%m%d%H%M')
             save_path = 'checkpoints/fasterrcnn_%s' % timestr
             for k_,v_ in kwargs.items():
                 save_path += '_%s' %v_