enable negative date train

README.md

@@ -1,3 +1,9 @@
+# Modify
+
+  this repo main difference from original is add negative data train
+
+  you can now add some image path without any annotions in you label.txt,then it can train now!
+
 # RetinaFace in PyTorch
 
 A [PyTorch](https://pytorch.org/) implementation of [RetinaFace: Single-stage Dense Face Localisation in the Wild](https://arxiv.org/abs/1905.00641). Model size only 1.7M, when Retinaface use mobilenet0.25 as backbone net. We also provide resnet50 as backbone net to get better result. The official code in Mxnet can be found [here](https://github.com/deepinsight/insightface/tree/master/RetinaFace).

data/wider_face.py

@@ -44,33 +44,55 @@ def __getitem__(self, index):
         labels = self.words[index]
         annotations = np.zeros((0, 15))
         if len(labels) == 0:
-            return annotations
-        for idx, label in enumerate(labels):
+            # return annotations
             annotation = np.zeros((1, 15))
             # bbox
-            annotation[0, 0] = label[0]  # x1
-            annotation[0, 1] = label[1]  # y1
-            annotation[0, 2] = label[0] + label[2]  # x2
-            annotation[0, 3] = label[1] + label[3]  # y2
+            annotation[0, 0] = 0  # x1
+            annotation[0, 1] = 0  # y1
+            annotation[0, 2] = 0  # x2
+            annotation[0, 3] = 0  # y2
 
             # landmarks
-            annotation[0, 4] = label[4]    # l0_x
-            annotation[0, 5] = label[5]    # l0_y
-            annotation[0, 6] = label[7]    # l1_x
-            annotation[0, 7] = label[8]    # l1_y
-            annotation[0, 8] = label[10]   # l2_x
-            annotation[0, 9] = label[11]   # l2_y
-            annotation[0, 10] = label[13]  # l3_x
-            annotation[0, 11] = label[14]  # l3_y
-            annotation[0, 12] = label[16]  # l4_x
-            annotation[0, 13] = label[17]  # l4_y
-            if (annotation[0, 4]<0):
-                annotation[0, 14] = -1
-            else:
-                annotation[0, 14] = 1
-
+            annotation[0, 4] = -1  # l0_x
+            annotation[0, 5] = -1   # l0_y
+            annotation[0, 6] = -1   # l1_x
+            annotation[0, 7] = -1   # l1_y
+            annotation[0, 8] = -1   # l2_x
+            annotation[0, 9] = -1   # l2_y
+            annotation[0, 10] = -1   # l3_x
+            annotation[0, 11] = -1   # l3_y
+            annotation[0, 12] = -1   # l4_x
+            annotation[0, 13] = -1   # l4_y
+            annotation[0, 14] = 0
             annotations = np.append(annotations, annotation, axis=0)
-        target = np.array(annotations)
+            target = np.array(annotations)
+        else:
+            for idx, label in enumerate(labels):
+                annotation = np.zeros((1, 15))
+                # bbox
+                annotation[0, 0] = label[0]  # x1
+                annotation[0, 1] = label[1]  # y1
+                annotation[0, 2] = label[0] + label[2]  # x2
+                annotation[0, 3] = label[1] + label[3]  # y2
+
+                # landmarks
+                annotation[0, 4] = label[4]    # l0_x
+                annotation[0, 5] = label[5]    # l0_y
+                annotation[0, 6] = label[7]    # l1_x
+                annotation[0, 7] = label[8]    # l1_y
+                annotation[0, 8] = label[10]   # l2_x
+                annotation[0, 9] = label[11]   # l2_y
+                annotation[0, 10] = label[13]  # l3_x
+                annotation[0, 11] = label[14]  # l3_y
+                annotation[0, 12] = label[16]  # l4_x
+                annotation[0, 13] = label[17]  # l4_y
+                if (annotation[0, 4]<0):
+                    annotation[0, 14] = -1
+                else:
+                    annotation[0, 14] = 1
+
+                annotations = np.append(annotations, annotation, axis=0)
+            target = np.array(annotations)
         if self.preproc is not None:
             img, target = self.preproc(img, target)
 

detect.py

@@ -23,7 +23,9 @@
 parser.add_argument('--nms_threshold', default=0.4, type=float, help='nms_threshold')
 parser.add_argument('--keep_top_k', default=750, type=int, help='keep_top_k')
 parser.add_argument('-s', '--save_image', action="store_true", default=True, help='show detection results')
-parser.add_argument('--vis_thres', default=0.6, type=float, help='visualization_threshold')
+parser.add_argument('--vis_thres', default=0.8, type=float, help='visualization_threshold')
+parser.add_argument('-img', '--image', default='./curve/test.jpg',
+                    type=str, help='image')
 args = parser.parse_args()
 
 
@@ -83,8 +85,8 @@ def load_model(model, pretrained_path, load_to_cpu):
     resize = 1
 
     # testing begin
-    for i in range(100):
-        image_path = "./curve/test.jpg"
+    for i in range(1):
+        image_path = args.image
         img_raw = cv2.imread(image_path, cv2.IMREAD_COLOR)
 
         img = np.float32(img_raw)
@@ -145,6 +147,7 @@ def load_model(model, pretrained_path, load_to_cpu):
         # show image
         if args.save_image:
             for b in dets:
+                print(b[0],b[1],b[2],b[3],b[4])
                 if b[4] < args.vis_thres:
                     continue
                 text = "{:.4f}".format(b[4])
@@ -164,5 +167,4 @@ def load_model(model, pretrained_path, load_to_cpu):
             # save image
 
             name = "test.jpg"
-            cv2.imwrite(name, img_raw)
-
+            cv2.imwrite(name, img_raw)

layers/modules/multibox_loss.py

@@ -106,7 +106,7 @@ def forward(self, predictions, priors, targets):
         _, loss_idx = loss_c.sort(1, descending=True)
         _, idx_rank = loss_idx.sort(1)
         num_pos = pos.long().sum(1, keepdim=True)
-        num_neg = torch.clamp(self.negpos_ratio*num_pos, max=pos.size(1)-1)
+        num_neg = torch.clamp(self.negpos_ratio*num_pos+200, max=pos.size(1)-1)#避免pos为0时，neg也为0，可以自行设置
         neg = idx_rank < num_neg.expand_as(idx_rank)
 
         # Confidence Loss Including Positive and Negative Examples