自定义数据集在计算loss时出错 about redet

[csuhan]

1. 使用低版本pytorch
   或者
2. #15

[zhangvia]

1. 使用低版本pytorch
   或者
2. #15

我两个方法都用了，会报同样的错，segmentation fault

[zhangvia]

我更新了gcc解决了segmentation fault 之后重新编译了mmdet，我修改了configs文件中的数据路径，修改datatype为cocodataset，修改了imagescale，numclasses，以及lr（因为我只有单张1080ti）。现在模型跑起来后，会自动终止什么信息也没有是怎么回事。。。运行train.py后信息如下：
[localchao@localhost ReDet]$ python tools/train.py configs/ReDet/ReDet_re50_refpn_1x_dota1.py
ReResNet Orientation: 8 Fix Params: False
2021-06-20 00:37:38,683 - INFO - Distributed training: False
2021-06-20 00:38:11,890 - INFO - load model from: work_dirs/ReResNet_pretrain/re_resnet50_c8_batch256-25b16846.pth
2021-06-20 00:38:11,940 - WARNING - The model and loaded state dict do not match exactly

unexpected key in source state_dict: head.fc.weight, head.fc.bias

missing keys in source state_dict: layer1.2.conv3.filter, layer3.1.conv1.filter, layer4.2.conv2.filter, layer3.1.conv3.filter, layer4.2.conv3.filter, layer1.1.conv1.filter, layer3.4.conv1.filter, layer4.2.conv1.filter, layer2.0.conv2.filter, layer1.2.conv2.filter, layer3.4.conv2.filter, conv1.filter, layer2.1.conv2.filter, layer3.5.conv1.filter, layer1.1.conv2.filter, layer2.3.conv1.filter, layer1.0.conv2.filter, layer2.3.conv2.filter, layer2.2.conv1.filter, layer2.2.conv2.filter, layer2.1.conv1.filter, layer3.2.conv1.filter, layer3.3.conv3.filter, layer1.2.conv1.filter, layer3.5.conv2.filter, layer3.1.conv2.filter, layer4.1.conv1.filter, layer4.0.conv3.filter, layer2.2.conv3.filter, layer3.0.downsample.0.filter, layer1.0.conv1.filter, layer2.3.conv3.filter, layer4.1.conv3.filter, layer3.2.conv2.filter, layer4.1.conv2.filter, layer3.0.conv2.filter, layer1.1.conv3.filter, layer3.4.conv3.filter, layer4.0.downsample.0.filter, layer3.3.conv1.filter, layer3.0.conv1.filter, layer2.0.downsample.0.filter, layer2.0.conv3.filter, layer3.5.conv3.filter, layer1.0.conv3.filter, layer3.0.conv3.filter, layer3.3.conv2.filter, layer2.1.conv3.filter, layer3.2.conv3.filter, layer4.0.conv1.filter, layer4.0.conv2.filter, layer2.0.conv1.filter, layer1.0.downsample.0.filter

loading annotations into memory...
Done (t=0.01s)
creating index...
index created!
2021-06-20 00:38:13,770 - INFO - Start running, host: [email protected], work_dir: /home/localchao/LHX/jyz/ReDet/work_dirs/ReDet_re50_refpn_1x_dota1
2021-06-20 00:38:13,770 - INFO - workflow: [('train', 1)], max: 12 epochs
[localchao@localhost ReDet]$
@csuhan

[csuhan]

可以在模型内forward_train加断点，查看是否有运行