predict.py测试出现问题

[Debugerss]

我想直接拿作者的predict函数来测试效果。但是在模型加载的时候出现模型key值不匹配的问题，怎么解决

[WenmuZhou]

能否贴下报错信息

[Debugerss]

RuntimeError: Error(s) in loading state_dict for PSENet:
Missing key(s) in state_dict: "toplayer.0.weight", "toplayer.0.bias", "toplayer.1.weight", "toplayer.1.bias", "toplayer.1.running_mean", "toplayer.1.running_var", "latlayer1.0.weight", "latlayer1.0.bias", "latlayer1.1.weight", "latlayer1.1.bias", "latlayer1.1.running_mean", "latlayer1.1.running_var", "latlayer2.0.weight", "latlayer2.0.bias", "latlayer2.1.weight", "latlayer2.1.bias", "latlayer2.1.running_mean", "latlayer2.1.running_var", "latlayer3.0.weight", "latlayer3.0.bias", "latlayer3.1.weight", "latlayer3.1.bias", "latlayer3.1.running_mean", "latlayer3.1.running_var", "smooth1.0.weight", "smooth1.0.bias", "smooth1.1.weight", "smooth1.1.bias", "smooth1.1.running_mean", "smooth1.1.running_var", "smooth2.0.weight", "smooth2.0.bias", "smooth2.1.weight", "smooth2.1.bias", "smooth2.1.running_mean", "smooth2.1.running_var", "smooth3.0.weight", "smooth3.0.bias", "smooth3.1.weight", "smooth3.1.bias", "smooth3.1.running_mean", "smooth3.1.running_var".
Unexpected key(s) in state_dict: "toplayer.weight", "toplayer.bias", "latlayer1.weight", "latlayer1.bias", "latlayer2.weight", "latlayer2.bias", "latlayer3.weight", "latlayer3.bias", "smooth1.weight", "smooth1.bias", "smooth2.weight", "smooth2.bias", "smooth3.weight", "smooth3.bias".

[Debugerss]

我用的是您网盘里的resnet50的模型，还请您指教

[WenmuZhou]

这个 这样的，网络结构我改了下，网盘里的模型就不能用了

[Debugerss]

哦，好吧，谢谢您耐心的回答
不知您是否可以上传一个可以预测的模型，我想测试一下环境是否有问题，方便后期训练

[WenmuZhou]

我手上目前没有可用的模型，不好意思

[Debugerss]

if label == '' or label == '###':
text_tags.append(True)
else:
text_tags.append(False)
您好，这句话的意思是，标签为“###”或者“”的加入训练，
否则就不加入训练，是吗？

[WenmuZhou]

不是，是相反的意思

[Debugerss]

您好，训练输入图像的尺寸是利用config文件里面的data_shape = 1280控制吗？
测试时输入图像的尺寸，默认2240是吗？ long_size: int = 2240):
想找hub主确认一下

[WenmuZhou]

是的

[Debugerss]

如果我设定data_shape = 1280，那输入图像的尺寸为什么是12801280
作者处理输入图像的代码是这个吗？
img, score_maps, training_mask = image_label(img_path, text_polys, text_tags, input_size=self.data_shape, n=self.n, m=self.m)
然后跳进image_label里面，
imgs = data_aug.random_crop_author([im, score_maps.transpose((1, 2, 0)),training_mask], (input_size, input_size))
我分析输入图像尺寸是12801280，是不是因为您在这里crop了一下

[Debugerss]

楼主，我在训练了一半之后，停电了，想继续训练，修改如下：
#checkpoint = '' checkpoint = '/dahuafs/userdata/01_PDF/01_PSENet/output/psenet_icd2015_resnet152_4gpu_author_crop_adam_MultiStepLR_authorloss/PSENet_36_loss0.055750.pth'
restart_training = False

#start_epoch = 0
start_epoch = 37
但是在训练的时候碰到，
PermissionError: [Errno 13] Permission denied: 'train_log'

请问怎么解决一下