Merge pull request WZMIAOMIAO#546 from WZMIAOMIAO/dev

reset weights path

Author: WZMIAOMIAO

pytorch_object_detection/faster_rcnn/train_res50_fpn.py

@@ -17,7 +17,7 @@ def create_model(num_classes, load_pretrain_weights=True):
     # 如果GPU显存很大可以设置比较大的batch_size就可以将norm_layer设置为普通的BatchNorm2d
     # trainable_layers包括['layer4', 'layer3', 'layer2', 'layer1', 'conv1']， 5代表全部训练
     # resnet50 imagenet weights url: https://download.pytorch.org/models/resnet50-0676ba61.pth
-    backbone = resnet50_fpn_backbone(pretrain_path="resnet50.pth",
+    backbone = resnet50_fpn_backbone(pretrain_path="./backbone/resnet50.pth",
                                      norm_layer=torch.nn.BatchNorm2d,
                                      trainable_layers=3)
     # 训练自己数据集时不要修改这里的91，修改的是传入的num_classes参数