fix the test bug

Author: lzx1413

data/config.py

@@ -2,12 +2,10 @@
 import os.path
 
 # gets home dir cross platform
-home = os.path.expanduser("~")
-ddir = os.path.join(home,"data/VOCdevkit/")
 
 # note: if you used our download scripts, this should be right
-VOCroot = ddir # path to VOCdevkit root dir
-COCOroot = os.path.join(home,"data/COCO/")
+VOCroot = '/mnt/lvmhdd1/zuoxin/dataset/VOCdevkit' # path to VOCdevkit root dir
+COCOroot = ''
 
 
 #RFB CONFIGS
@@ -32,7 +30,7 @@
 VOC_512= {
     'feature_maps' : [64, 32, 16, 8, 4, 2, 1],
 
-    'min_dim' : 524,
+    'min_dim' : 512,
 
     'steps' : [8, 16, 32, 64, 128, 256, 512],
 

data/voc0712.py

@@ -250,7 +250,8 @@ def evaluate_detections(self, all_boxes, output_dir=None):
         all_boxes[class][image] = [] or np.array of shape #dets x 5
         """
         self._write_voc_results_file(all_boxes)
-        self._do_python_eval(output_dir)
+        aps,map = self._do_python_eval(output_dir)
+        return aps,map
 
     def _get_voc_results_file_template(self):
         filename = 'comp4_det_test' + '_{:s}.txt'
@@ -327,6 +328,7 @@ def _do_python_eval(self, output_dir='output'):
         print('Recompute with `./tools/reval.py --matlab ...` for your paper.')
         print('-- Thanks, The Management')
         print('--------------------------------------------------------------')
+        return aps,np.mean(aps)
 
 def detection_collate(batch):
     """Custom collate fn for dealing with batches of images that have a different

models/RFB_Net_E_vgg.py

@@ -185,9 +185,8 @@ def forward(self,x):
 
 class RFBNet(nn.Module):
 
-    def __init__(self, phase, size, base, extras, head, num_classes):
+    def __init__(self,  size, base, extras, head, num_classes):
         super(RFBNet, self).__init__()
-        self.phase = phase
         self.num_classes = num_classes
         self.size = size
 
@@ -209,10 +208,9 @@ def __init__(self, phase, size, base, extras, head, num_classes):
 
         self.loc = nn.ModuleList(head[0])
         self.conf = nn.ModuleList(head[1])
-        if self.phase == 'test':
-            self.softmax = nn.Softmax()
+        self.softmax = nn.Softmax()
 
-    def forward(self, x):
+    def forward(self, x,test=False):
         """Applies network layers and ops on input image(s) x.
 
         Args:
@@ -268,7 +266,7 @@ def forward(self, x):
         loc = torch.cat([o.view(o.size(0), -1) for o in loc], 1)
         conf = torch.cat([o.view(o.size(0), -1) for o in conf], 1)
 
-        if self.phase == "test":
+        if test:
             output = (
                 loc.view(loc.size(0), -1, 4),                   # loc preds
                 self.softmax(conf.view(-1, self.num_classes)),  # conf preds
@@ -368,14 +366,11 @@ def multibox(size, vgg, extra_layers, cfg, num_classes):
 }
 
 
-def build_net(phase, size=300, num_classes=21):
-    if phase != "test" and phase != "train":
-        print("Error: Phase not recognized")
-        return
+def build_net(size=300, num_classes=21):
     if size != 300 and size != 512:
         print("Error: Sorry only RFB300 and RFB512 are supported!")
         return
 
-    return RFBNet(phase, size, *multibox(size, vgg(vgg_base[str(size)], 3),
+    return RFBNet(size, *multibox(size, vgg(vgg_base[str(size)], 3),
                                 add_extras(size, extras[str(size)], 1024),
-                                mbox[str(size)], num_classes), num_classes)
+                                mbox[str(size)], num_classes), num_classes=num_classes)

models/RFB_Net_mobile.py

@@ -154,10 +154,9 @@ def __init__(self, phase, size, base, extras, head, num_classes):
 
         self.loc = nn.ModuleList(head[0])
         self.conf = nn.ModuleList(head[1])
-        if self.phase == 'test':
-            self.softmax = nn.Softmax()
+        self.softmax = nn.Softmax()
 
-    def forward(self, x):
+    def forward(self, x,test=False):
         """Applies network layers and ops on input image(s) x.
 
         Args:
@@ -209,7 +208,7 @@ def forward(self, x):
         loc = torch.cat([o.view(o.size(0), -1) for o in loc], 1)
         conf = torch.cat([o.view(o.size(0), -1) for o in conf], 1)
 
-        if self.phase == "test":
+        if test:
             output = (
                 loc.view(loc.size(0), -1, 4),                   # loc preds
                 self.softmax(conf.view(-1, self.num_classes)),  # conf preds
@@ -335,14 +334,11 @@ def multibox(size, base, extra_layers, cfg, num_classes):
 }
 
 
-def build_net(phase, size=300, num_classes=21):
-    if phase != "test" and phase != "train":
-        print("Error: Phase not recognized")
-        return
+def build_net(size=300, num_classes=21):
     if size != 300:
         print("Error: Sorry only RFB300_mobile is supported!")
         return
 
-    return RFBNet(phase, size, *multibox(size, MobileNet(),
+    return RFBNet(size, *multibox(size, MobileNet(),
                                 add_extras(size, extras[str(size)], 1024),
-                                mbox[str(size)], num_classes), num_classes)
+                                mbox[str(size)], num_classes), num_classes=num_classes)

models/RFB_Net_vgg.py

@@ -138,9 +138,8 @@ class RFBNet(nn.Module):
         head: "multibox head" consists of loc and conf conv layers
     """
 
-    def __init__(self, phase, size, base, extras, head, num_classes):
+    def __init__(self,size, base, extras, head, num_classes):
         super(RFBNet, self).__init__()
-        self.phase = phase
         self.num_classes = num_classes
         self.size = size
 
@@ -159,10 +158,9 @@ def __init__(self, phase, size, base, extras, head, num_classes):
 
         self.loc = nn.ModuleList(head[0])
         self.conf = nn.ModuleList(head[1])
-        if self.phase == 'test':
-            self.softmax = nn.Softmax()
+        self.softmax = nn.Softmax()
 
-    def forward(self, x):
+    def forward(self, x, test = False):
         """Applies network layers and ops on input image(s) x.
 
         Args:
@@ -214,7 +212,7 @@ def forward(self, x):
         loc = torch.cat([o.view(o.size(0), -1) for o in loc], 1)
         conf = torch.cat([o.view(o.size(0), -1) for o in conf], 1)
 
-        if self.phase == "test":
+        if test:
             output = (
                 loc.view(loc.size(0), -1, 4),                   # loc preds
                 self.softmax(conf.view(-1, self.num_classes)),  # conf preds
@@ -311,14 +309,11 @@ def multibox(size, vgg, extra_layers, cfg, num_classes):
 }
 
 
-def build_net(phase, size=300, num_classes=21):
-    if phase != "test" and phase != "train":
-        print("Error: Phase not recognized")
-        return
+def build_net(size=300, num_classes=21):
     if size != 300 and size != 512:
         print("Error: Sorry only RFBNet300 and RFBNet512 are supported!")
         return
 
-    return RFBNet(phase, size, *multibox(size, vgg(vgg_base[str(size)], 3),
+    return RFBNet(size, *multibox(size, vgg(vgg_base[str(size)], 3),
                                 add_extras(size, extras[str(size)], 1024),
-                                mbox[str(size)], num_classes), num_classes)
+                                mbox[str(size)], num_classes), num_classes=num_classes)

models/SSD_vgg.py

@@ -23,9 +23,8 @@ class SSD(nn.Module):
         head: "multibox head" consists of loc and conf conv layers
     """
 
-    def __init__(self, phase, base, extras, head, num_classes):
+    def __init__(self,base, extras, head, num_classes):
         super(SSD, self).__init__()
-        self.phase = phase
         self.num_classes = num_classes
         # TODO: implement __call__ in PriorBox
         self.size = 300
@@ -39,11 +38,9 @@ def __init__(self, phase, base, extras, head, num_classes):
         self.loc = nn.ModuleList(head[0])
         self.conf = nn.ModuleList(head[1])
 
-        if phase == 'test':
-            self.softmax = nn.Softmax()
-            self.detect = Detect(num_classes, 0, 200, 0.01, 0.45)
+        self.softmax = nn.Softmax()
 
-    def forward(self, x):
+    def forward(self, x,test=False):
         """Applies network layers and ops on input image(s) x.
 
         Args:
@@ -91,8 +88,8 @@ def forward(self, x):
 
         loc = torch.cat([o.view(o.size(0), -1) for o in loc], 1)
         conf = torch.cat([o.view(o.size(0), -1) for o in conf], 1)
-        if self.phase == "test":
-            output = self.detect(
+        if test:
+            output =(
                 loc.view(loc.size(0), -1, 4),                   # loc preds
                 self.softmax(conf.view(-1, self.num_classes)),  # conf preds
             )
@@ -159,14 +156,11 @@ def multibox(vgg, extra_layers, cfg, num_classes):
 }
 
 
-def build_net(phase, size=300, num_classes=21):
-    if phase != "test" and phase != "train":
-        print("Error: Phase not recognized")
-        return
+def build_net(size=300, num_classes=21):
     if size != 300 and size != 512:
         print("Error: Sorry only SSD300 and SSD512 is supported currently!")
         return
 
-    return SSD(phase, *multibox(vgg(vgg_base[str(size)], 3),
+    return SSD(*multibox(vgg(vgg_base[str(size)], 3),
                                 add_extras(extras[str(size)], 1024),
-                                mbox[str(size)], num_classes), num_classes)
+                                mbox[str(size)], num_classes), num_classes=num_classes)

test_RFB.py

@@ -85,7 +85,7 @@ def test_net(save_folder, net, detector, cuda, testset, transform, max_per_image
             x = x.cuda()
 
         _t['im_detect'].tic()
-        out = net(x)      # forward pass
+        out = net(x,test = True)      # forward pass
         boxes, scores = detector.forward(out,priors)
         detect_time = _t['im_detect'].toc()
         boxes = boxes[0]
@@ -145,7 +145,7 @@ def test_net(save_folder, net, detector, cuda, testset, transform, max_per_image
     # load net
     img_dim = (300,512)[args.size=='512']
     num_classes = (21, 81)[args.dataset == 'COCO']
-    net = build_net('test', img_dim, num_classes)    # initialize detector
+    net = build_net(img_dim, num_classes)    # initialize detector
     state_dict = torch.load(args.trained_model)
     # create new OrderedDict that does not contain `module.`
 

train_test.py

@@ -25,7 +25,7 @@ def str2bool(v):
 parser = argparse.ArgumentParser(
     description='Receptive Field Block Net Training')
 parser.add_argument('-v', '--version', default='SSD_vgg',
-                    help='RFB_vgg ,RFB_E_vgg RFB_mobile SSD version.')
+                    help='RFB_vgg ,RFB_E_vgg RFB_mobile SSD_vgg version.')
 parser.add_argument('-s', '--size', default='300',
                     help='300 or 512 input size.')
 parser.add_argument('-d', '--dataset', default='VOC',
@@ -45,8 +45,8 @@ def str2bool(v):
                     default=1e-3, type=float, help='initial learning rate')
 parser.add_argument('--momentum', default=0.9, type=float, help='momentum')
 parser.add_argument(
-    '--resume_net', default=False, help='resume net for retraining')
-parser.add_argument('--resume_epoch', default=0,
+    '--resume_net', default=True, help='resume net for retraining')
+parser.add_argument('--resume_epoch', default=250,
                     type=int, help='resume iter for retraining')
 parser.add_argument('-max','--max_epoch', default=300,
                     type=int, help='max epoch for retraining')
@@ -62,8 +62,8 @@ def str2bool(v):
 parser.add_argument('--save_frequency',default=10)
 parser.add_argument('--retest', default=False, type=bool,
                     help='test cache results')
-parser.add_argument('--test_frequency',default=100)
-parser.add_argument('--visdom', default=True, type=str2bool, help='Use visdom to for loss visualization')
+parser.add_argument('--test_frequency',default=10)
+parser.add_argument('--visdom', default=False, type=str2bool, help='Use visdom to for loss visualization')
 parser.add_argument('--send_images_to_visdom', type=str2bool, default=False, help='Sample a random image from each 10th batch, send it to visdom after augmentations step')
 args = parser.parse_args()
 
@@ -110,7 +110,7 @@ def str2bool(v):
     import visdom
     viz = visdom.Visdom()
 
-net = build_net('train', img_dim, num_classes)
+net = build_net(img_dim, num_classes)
 print(net)
 if not args.resume_net:
     base_weights = torch.load(args.basenet)
@@ -318,7 +318,7 @@ def train():
                 win=lot,
                 update='append'
             )
-            if iteration == 0:
+            if iteration%epoch_size == 0:
                 viz.line(
                     X=torch.zeros((1, 3)).cpu(),
                     Y=torch.Tensor([loc_loss, conf_loss,
@@ -371,7 +371,7 @@ def test_net(save_folder, net, detector, cuda, testset, transform, max_per_image
             x = x.cuda()
 
         _t['im_detect'].tic()
-        out = net(x)      # forward pass
+        out = net(x = x,test = True)      # forward pass
         boxes, scores = detector.forward(out,priors)
         detect_time = _t['im_detect'].toc()
         boxes = boxes[0]
@@ -424,7 +424,10 @@ def test_net(save_folder, net, detector, cuda, testset, transform, max_per_image
         pickle.dump(all_boxes, f, pickle.HIGHEST_PROTOCOL)
 
     print('Evaluating detections')
-    testset.evaluate_detections(all_boxes, save_folder)
+    if args.dataset == 'VOC':
+        aps,map = testset.evaluate_detections(all_boxes, save_folder)
+        return aps,map
+
 
 
 if __name__ == '__main__':