support Pytorch 0.4

fengfengfeng96 · Jul 26, 2018 · 0f5d9aa · 0f5d9aa
1 parent 6544e53
commit 0f5d9aa
Show file tree

Hide file tree

Showing 45 changed files with 38 additions and 45 deletions.
diff --git a/.test_RFB.py.swp b/.test_RFB.py.swp
diff --git a/README.md b/README.md
@@ -5,7 +5,7 @@ By Songtao Liu, Di Huang, Yunhong Wang
 ### Introduction
 Inspired by the structure of Receptive Fields (RFs) in human visual systems, we propose a novel RF Block (RFB) module, which takes the relationship between the size and eccentricity of RFs into account, to enhance the discriminability and robustness of features. We further  assemble the RFB module to the top of SSD with a lightweight CNN model, constructing the RFB Net detector. You can use the code to train/evaluate the RFB Net for object detection. For more details, please refer to our [arXiv paper](https://arxiv.org/pdf/1711.07767.pdf). 
 
-<img align="right" src="https://github.com/ruinmessi/RFBNet/blob/master/doc/rfb.png">
+<img align="right" src="https://github.com/ruinmessi/RFBNet/blob/master/doc/RFB.png">
 
 &nbsp;
 &nbsp;

diff --git a/data/__pycache__/__init__.cpython-36.pyc b/data/__pycache__/__init__.cpython-36.pyc
diff --git a/data/__pycache__/coco.cpython-36.pyc b/data/__pycache__/coco.cpython-36.pyc
diff --git a/data/__pycache__/config.cpython-36.pyc b/data/__pycache__/config.cpython-36.pyc
diff --git a/data/__pycache__/data_augment.cpython-36.pyc b/data/__pycache__/data_augment.cpython-36.pyc
diff --git a/data/__pycache__/voc0712.cpython-36.pyc b/data/__pycache__/voc0712.cpython-36.pyc
diff --git a/data/__pycache__/voc_eval.cpython-36.pyc b/data/__pycache__/voc_eval.cpython-36.pyc
diff --git a/data/data_augment.py b/data/data_augment.py
@@ -3,10 +3,6 @@
 
 The data augmentation procedures were interpreted from @weiliu89's SSD paper
 http://arxiv.org/abs/1512.02325
-
-TODO: implement data_augment for training
-
-Ellis Brown, Max deGroot
 """
 
 import torch

diff --git a/doc/RFB.png b/doc/RFB.png
diff --git a/doc/rfb.png b/doc/rfb.png
diff --git a/layers/__pycache__/__init__.cpython-36.pyc b/layers/__pycache__/__init__.cpython-36.pyc
diff --git a/layers/functions/__pycache__/__init__.cpython-36.pyc b/layers/functions/__pycache__/__init__.cpython-36.pyc
diff --git a/layers/functions/__pycache__/detection.cpython-36.pyc b/layers/functions/__pycache__/detection.cpython-36.pyc
diff --git a/layers/functions/__pycache__/prior_box.cpython-36.pyc b/layers/functions/__pycache__/prior_box.cpython-36.pyc
diff --git a/layers/functions/detection.py b/layers/functions/detection.py
@@ -3,7 +3,7 @@
 import torch.backends.cudnn as cudnn
 from torch.autograd import Function
 from torch.autograd import Variable
-from utils.box_utils import decode, nms
+from utils.box_utils import decode
 
 
 class Detect(Function):
@@ -15,9 +15,7 @@ class Detect(Function):
     def __init__(self, num_classes, bkg_label, cfg):
         self.num_classes = num_classes
         self.background_label = bkg_label
-        #self.thresh = thresh
 
-        # Parameters used in nms.
         self.variance = cfg['variance']
 
     def forward(self, predictions, prior):
@@ -40,6 +38,9 @@ def forward(self, predictions, prior):
         self.num_priors = prior_data.size(0)
         self.boxes = torch.zeros(1, self.num_priors, 4)
         self.scores = torch.zeros(1, self.num_priors, self.num_classes)
+        if loc_data.is_cuda:
+            self.boxes = self.boxes.cuda()
+            self.scores = self.scores.cuda()
 
         if num == 1:
             # size batch x num_classes x num_priors
@@ -54,13 +55,7 @@ def forward(self, predictions, prior):
         # Decode predictions into bboxes.
         for i in range(num):
             decoded_boxes = decode(loc_data[i], prior_data, self.variance)
-            # For each class, perform nms
             conf_scores = conf_preds[i].clone()
-            '''
-            c_mask = conf_scores.gt(self.thresh)
-            decoded_boxes = decoded_boxes[c_mask]
-            conf_scores = conf_scores[c_mask]
-            '''
 
             self.boxes[i] = decoded_boxes
             self.scores[i] = conf_scores

diff --git a/layers/functions/prior_box.py b/layers/functions/prior_box.py
@@ -3,8 +3,6 @@
 import torch.backends.cudnn as cudnn
 from math import sqrt as sqrt
 from itertools import product as product
-if torch.cuda.is_available():
-    torch.set_default_tensor_type('torch.cuda.FloatTensor')
 
 
 class PriorBox(object):

diff --git a/layers/modules/__pycache__/__init__.cpython-36.pyc b/layers/modules/__pycache__/__init__.cpython-36.pyc
diff --git a/layers/modules/__pycache__/multibox_loss.cpython-36.pyc b/layers/modules/__pycache__/multibox_loss.cpython-36.pyc
diff --git a/layers/modules/multibox_loss.py b/layers/modules/multibox_loss.py
@@ -6,7 +6,6 @@
 GPU = False
 if torch.cuda.is_available():
     GPU = True
-    torch.set_default_tensor_type('torch.cuda.FloatTensor')
 
 
 class MultiBoxLoss(nn.Module):
@@ -93,7 +92,7 @@ def forward(self, predictions, priors, targets):
         loss_c = log_sum_exp(batch_conf) - batch_conf.gather(1, conf_t.view(-1,1))
 
         # Hard Negative Mining
-        loss_c[pos.view(-1)] = 0 # filter out pos boxes for now
+        loss_c[pos.view(-1,1)] = 0 # filter out pos boxes for now
         loss_c = loss_c.view(num, -1)
         _,loss_idx = loss_c.sort(1, descending=True)
         _,idx_rank = loss_idx.sort(1)
@@ -110,7 +109,7 @@ def forward(self, predictions, priors, targets):
 
         # Sum of losses: L(x,c,l,g) = (Lconf(x, c) + αLloc(x,l,g)) / N
 
-        N = num_pos.data.sum()
+        N = max(num_pos.data.sum().float(), 1)
         loss_l/=N
         loss_c/=N
         return loss_l,loss_c
diff --git a/models/RFB_Net_E_vgg.py b/models/RFB_Net_E_vgg.py
@@ -208,7 +208,7 @@ 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(dim=-1)
 
     def forward(self, x):
         """Applies network layers and ops on input image(s) x.

diff --git a/models/RFB_Net_mobile.py b/models/RFB_Net_mobile.py
@@ -155,7 +155,7 @@ 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(dim=-1)
 
     def forward(self, x):
         """Applies network layers and ops on input image(s) x.

diff --git a/models/RFB_Net_vgg.py b/models/RFB_Net_vgg.py
@@ -158,7 +158,7 @@ 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(dim=-1)
 
     def forward(self, x):
         """Applies network layers and ops on input image(s) x.

diff --git a/models/__pycache__/RFB_Net_vgg.cpython-36.pyc b/models/__pycache__/RFB_Net_vgg.cpython-36.pyc
diff --git a/models/__pycache__/__init__.cpython-36.pyc b/models/__pycache__/__init__.cpython-36.pyc
diff --git a/test_RFB.py b/test_RFB.py
@@ -31,6 +31,8 @@
                     help='Dir to save results')
 parser.add_argument('--cuda', default=True, type=bool,
                     help='Use cuda to train model')
+parser.add_argument('--cpu', default=False, type=bool,
+                    help='Use cpu nms')
 parser.add_argument('--retest', default=False, type=bool,
                     help='test cache results')
 args = parser.parse_args()
@@ -54,9 +56,10 @@
     print('Unkown version!')
 
 priorbox = PriorBox(cfg)
-priors = Variable(priorbox.forward(), volatile=True)
-if not args.cuda:
-    priors = priors.cpu()
+with torch.no_grad():
+    priors = priorbox.forward()
+    if args.cuda:
+        priors = priors.cuda()
 
 
 def test_net(save_folder, net, detector, cuda, testset, transform, max_per_image=300, thresh=0.005):
@@ -82,9 +85,13 @@ def test_net(save_folder, net, detector, cuda, testset, transform, max_per_image
 
     for i in range(num_images):
         img = testset.pull_image(i)
-        x = Variable(transform(img).unsqueeze(0),volatile=True)
-        if cuda:
-            x = x.cuda()
+        scale = torch.Tensor([img.shape[1], img.shape[0],
+                             img.shape[1], img.shape[0]])
+        with torch.no_grad():
+            x = transform(img).unsqueeze(0)
+            if cuda:
+                x = x.cuda()
+                scale = scale.cuda()
 
         _t['im_detect'].tic()
         out = net(x)      # forward pass
@@ -93,12 +100,10 @@ def test_net(save_folder, net, detector, cuda, testset, transform, max_per_image
         boxes = boxes[0]
         scores=scores[0]
 
+        boxes *= scale
         boxes = boxes.cpu().numpy()
         scores = scores.cpu().numpy()
         # scale each detection back up to the image
-        scale = torch.Tensor([img.shape[1], img.shape[0],
-                             img.shape[1], img.shape[0]]).cpu().numpy()
-        boxes *= scale
 
         _t['misc'].tic()
 
@@ -111,13 +116,8 @@ def test_net(save_folder, net, detector, cuda, testset, transform, max_per_image
             c_scores = scores[inds, j]
             c_dets = np.hstack((c_bboxes, c_scores[:, np.newaxis])).astype(
                 np.float32, copy=False)
-            if args.dataset == 'VOC':
-                cpu = True
-            else:
-                cpu = False
 
-            keep = nms(c_dets, 0.45, force_cpu=cpu)
-            keep = keep[:50]
+            keep = nms(c_dets, 0.45, force_cpu=args.cpu)
             c_dets = c_dets[keep, :]
             all_boxes[j][i] = c_dets
         if max_per_image > 0:

diff --git a/train_RFB.py b/train_RFB.py
@@ -99,7 +99,7 @@ def weights_init(m):
         for key in m.state_dict():
             if key.split('.')[-1] == 'weight':
                 if 'conv' in key:
-                    init.kaiming_normal(m.state_dict()[key], mode='fan_out')
+                    init.kaiming_normal_(m.state_dict()[key], mode='fan_out')
                 if 'bn' in key:
                     m.state_dict()[key][...] = 1
             elif key.split('.')[-1] == 'bias':
@@ -146,7 +146,11 @@ def weights_init(m):
 
 criterion = MultiBoxLoss(num_classes, 0.5, True, 0, True, 3, 0.5, False)
 priorbox = PriorBox(cfg)
-priors = Variable(priorbox.forward(), volatile=True)
+with torch.no_grad():
+    priors = priorbox.forward()
+    if args.cuda:
+        priors = priors.cuda()
+
 
 
 def train():
@@ -207,10 +211,10 @@ def train():
 
         if args.cuda:
             images = Variable(images.cuda())
-            targets = [Variable(anno.cuda(),volatile=True) for anno in targets]
+            targets = [Variable(anno.cuda()) for anno in targets]
         else:
             images = Variable(images)
-            targets = [Variable(anno, volatile=True) for anno in targets]
+            targets = [Variable(anno) for anno in targets]
         # forward
         t0 = time.time()
         out = net(images)
@@ -221,14 +225,14 @@ def train():
         loss.backward()
         optimizer.step()
         t1 = time.time()
-        loc_loss += loss_l.data[0]
-        conf_loss += loss_c.data[0]
+        loc_loss += loss_l.item()
+        conf_loss += loss_c.item()
         load_t1 = time.time()
         if iteration % 10 == 0:
             print('Epoch:' + repr(epoch) + ' || epochiter: ' + repr(iteration % epoch_size) + '/' + repr(epoch_size)
                   + '|| Totel iter ' +
                   repr(iteration) + ' || L: %.4f C: %.4f||' % (
-                loss_l.data[0],loss_c.data[0]) + 
+                loss_l.item(),loss_c.item()) + 
                 'Batch time: %.4f sec. ||' % (load_t1 - load_t0) + 'LR: %.8f' % (lr))
 
     torch.save(net.state_dict(), args.save_folder +

diff --git a/utils/__pycache__/__init__.cpython-36.pyc b/utils/__pycache__/__init__.cpython-36.pyc
diff --git a/utils/__pycache__/box_utils.cpython-36.pyc b/utils/__pycache__/box_utils.cpython-36.pyc
diff --git a/utils/__pycache__/nms_wrapper.cpython-36.pyc b/utils/__pycache__/nms_wrapper.cpython-36.pyc
diff --git a/utils/__pycache__/timer.cpython-36.pyc b/utils/__pycache__/timer.cpython-36.pyc
diff --git a/utils/box_utils.py b/utils/box_utils.py
@@ -4,7 +4,6 @@
 import numpy as np
 if torch.cuda.is_available():
     import torch.backends.cudnn as cudnn
-    torch.set_default_tensor_type('torch.cuda.FloatTensor')
 
 
 def point_form(boxes):
@@ -296,3 +295,5 @@ def nms(boxes, scores, overlap=0.5, top_k=200):
         # keep only elements with an IoU <= overlap
         idx = idx[IoU.le(overlap)]
     return keep, count
+
+
diff --git a/utils/build/temp.linux-x86_64-3.6/nms/cpu_nms.o b/utils/build/temp.linux-x86_64-3.6/nms/cpu_nms.o
diff --git a/utils/build/temp.linux-x86_64-3.6/nms/gpu_nms.o b/utils/build/temp.linux-x86_64-3.6/nms/gpu_nms.o
diff --git a/utils/build/temp.linux-x86_64-3.6/nms/nms_kernel.o b/utils/build/temp.linux-x86_64-3.6/nms/nms_kernel.o
diff --git a/utils/build/temp.linux-x86_64-3.6/pycocotools/_mask.o b/utils/build/temp.linux-x86_64-3.6/pycocotools/_mask.o
diff --git a/utils/build/temp.linux-x86_64-3.6/pycocotools/maskApi.o b/utils/build/temp.linux-x86_64-3.6/pycocotools/maskApi.o
diff --git a/utils/nms/__pycache__/__init__.cpython-36.pyc b/utils/nms/__pycache__/__init__.cpython-36.pyc
diff --git a/utils/nms/cpu_nms.cpython-36m-x86_64-linux-gnu.so b/utils/nms/cpu_nms.cpython-36m-x86_64-linux-gnu.so
diff --git a/utils/nms/gpu_nms.cpython-36m-x86_64-linux-gnu.so b/utils/nms/gpu_nms.cpython-36m-x86_64-linux-gnu.so
diff --git a/utils/pycocotools/__pycache__/__init__.cpython-36.pyc b/utils/pycocotools/__pycache__/__init__.cpython-36.pyc
diff --git a/utils/pycocotools/__pycache__/coco.cpython-36.pyc b/utils/pycocotools/__pycache__/coco.cpython-36.pyc
diff --git a/utils/pycocotools/__pycache__/cocoeval.cpython-36.pyc b/utils/pycocotools/__pycache__/cocoeval.cpython-36.pyc
diff --git a/utils/pycocotools/__pycache__/mask.cpython-36.pyc b/utils/pycocotools/__pycache__/mask.cpython-36.pyc
diff --git a/utils/pycocotools/_mask.cpython-36m-x86_64-linux-gnu.so b/utils/pycocotools/_mask.cpython-36m-x86_64-linux-gnu.so