No such file or directory: './saved_model/0816_adddice/wingsnet_37.pth

[GuobinZhangTJU]

您好，

非常冒昧打扰您，按照您在Github上的描述，timi团队代码中，Use file train2.py and train.py to train and test the model.

在运行train2.py时发现，当提取训练数据时，需要运行下面这行语句，

bp_weight = np.load(os.path.join(self.root, 'LIBBP', 'distance_bp', name + '.npy'))

可是其中的【'LIBBP', 'distance_bp', name + '.npy'】，并没有找到相关获取程序。 我往上翻了一下，如果我要得到'distance_bp',路径下面的文件就需要运行save_BP_weight_tw():， 但是save_BP_weight_tw():中又需要运行 pred = nibabel.load(os.path.join('./data/LIBBP/preds', name+'.nii.gz')) 然后要得到这个语句里面的/preds就需要运行save_gradients_tw(layer=0) 但是运行save_gradients_tw(layer=0)就需要运行weights_dict = torch.load(os.path.join('./saved_model', 'wingsnet_4.pth')) 可是当前我们是训练过程，是没法得到这个save_model的，不是训练结束后才能得到model么？ 您看可不可以帮忙解答一下~非常感谢您~

[Puzzled-Hui]

Hi, Thank you for your question.

In Timi team, they adopted a three stage method.

The loaded weight is the result of the first stage. Please refer to the challenge paper: for more details, especially the method section.

Best,

[GuobinZhangTJU]

Thanks for your reply. I have read the paper four times, but found no training steps, can you help me with the running steps? I have tried the following steps: LIB weight.py——save_gradients.py——BP enhancement.py——tree_parse.py——train2.py, the above step i think is reasonable, however i really not find the "weights_dict = torch.load(os.path.join('./saved_model', 'wingsnet_4.pth'))" in save_gradients.py waiting for your reply ~ thanks~

[KKoml]

您好，我也是没有save-model，请问您解决了吗？

[Puzzled-Hui]

Hi,

The loaded weight is the result of the first stage, which can be trained locally, and the wingsnet_4.pth is the early epoch results.

Best,

[TangHuiling]

您好，您上面提到wingsnet_4.pth是第一阶段训练的早期结果，但是在第一阶段的训练中，train.py文件中有一行 weights_dict = torch.load(os.path.join('./saved_model', 'wingsnet_79.pth'))， wingsnet_79.pth这个预训练权重文件并没有提供。请问可以提供一下吗？

[Puzzled-Hui]

@GuobinZhangTJU @KKoml @TangHuiling

Hi, we will deal with it and update the pretrained weights soon, thank you for your patience.

Best,

[Puzzled-Hui]

Dear all,

We have uploaded the pretrained weight: here.

Alternatively, you could pretrain the weight as the first stage use only Dice loss by yourselves. The early epochs' weight can be used for LIB weight, and the late epochs' weight can be used as the initial weight for the second stage.

Best,

[GuobinZhangTJU]

Dear all,

We have uploaded the pretrained weight: here.

Alternatively, you could pretrain the weight as the first stage use only Dice loss by yourselves. The early epochs' weight can be used for LIB weight, and the late epochs' weight can be used as the initial weight for the second stage.

Best,

Thanks for your reply~ best~