train error #44

smacaijicoder · 2021-12-31T15:56:02Z

执行nnFormer_train 3d_fullres nnFormerTrainerV2_Synapse 2 0命令时报错KeyError: 'img0005'
是为什么呢？数据预处理以安装说明执行nnFormer_plan_and_preprocess -t 2
改为nnFormer_train 3d_fullres nnFormerTrainerV2 2 0即可run

282857341 · 2021-12-31T16:06:01Z

nnFormerTrainerV2_Synapse 在这里设置了训练集和验证集的key，有问题应该是这里有问题
而nnFormerTrainerV2没有这一项

nnFormer/nnformer/training/network_training/nnFormerTrainerV2_Synapse.py

Lines 311 to 314 in e20bd3b

    
           splits[self.fold]['train']=np.array(['img0006','img0007' ,'img0009', 'img0010', 'img0021' ,'img0023' ,'img0024','img0026' ,'img0027' ,'img0031', 'img0033' ,'img0034' \ 
        
                               ,'img0039', 'img0040','img0005', 'img0028', 'img0030', 'img0037']) 
        
           splits[self.fold]['val']=np.array(['img0001', 'img0002', 'img0003', 'img0004', 'img0008', 'img0022','img0025', 'img0029', 'img0032', 'img0035', 'img0036', 'img0038'])

smacaijicoder · 2021-12-31T16:06:24Z

已收到邮件，将尽快回复！

282857341 · 2021-12-31T16:09:12Z

有可能是因为imagesTr 里没有img0005这一项

smacaijicoder · 2021-12-31T16:15:20Z

感谢回复，祝您元旦快乐！ imageTr中是有005的，难道是命名格式的问题吗？可我确实是按照nnunet要求的格式呀，而且用nnFormerTrainerV2是没有报错的 Best Wishes!  

…

------------------ 原始邮件 ------------------ 发件人: "282857341/nnFormer" ***@***.***>; 发送时间: 2022年1月1日(星期六) 凌晨0:09 ***@***.***>; ***@***.******@***.***>; 主题: Re: [282857341/nnFormer] train error (Issue #44) 有可能是因为imagesTr 里没有img0005这一项 — Reply to this email directly, view it on GitHub, or unsubscribe. Triage notifications on the go with GitHub Mobile for iOS or Android. You are receiving this because you authored the thread.Message ID: ***@***.***>

282857341 · 2021-12-31T16:19:14Z

能发一下报错的完整截图么

smacaijicoder · 2021-12-31T16:23:04Z

可以的，非常抱歉没有看到第一封邮件！确实如您所言，报错信息提示就在311-314行，此处划分难道是需要我根据自己的任务重新写吗 Best Wishes!  

…

------------------ 原始邮件 ------------------ 发件人: "282857341/nnFormer" ***@***.***>; 发送时间: 2022年1月1日(星期六) 凌晨0:19 ***@***.***>; ***@***.******@***.***>; 主题: Re: [282857341/nnFormer] train error (Issue #44) 能发一下报错的完整截图么 — Reply to this email directly, view it on GitHub, or unsubscribe. Triage notifications on the go with GitHub Mobile for iOS or Android. You are receiving this because you authored the thread.Message ID: ***@***.***>

smacaijicoder · 2021-12-31T16:32:04Z

issue似乎不能显示图片吗？
报错信息如下：
Traceback (most recent call last):
File "/home/yiqingwen/anaconda3/envs/nnFormer/bin/nnFormer_train", line 33, in
sys.exit(load_entry_point('nnformer', 'console_scripts', 'nnFormer_train')())
File "/home/yiqingwen/nnformer/nnFormer/nnformer/run/run_training.py", line 165, in main
trainer.initialize(not validation_only)
File "/home/yiqingwen/nnformer/nnFormer/nnformer/training/network_training/nnFormerTrainerV2_Synapse.py", line 96, in initialize
self.dl_tr, self.dl_val = self.get_basic_generators()
File "/home/yiqingwen/nnformer/nnFormer/nnformer/training/network_training/nnFormerTrainer_synapse.py", line 401, in get_basic_generators
self.do_split()
File "/home/yiqingwen/nnformer/nnFormer/nnformer/training/network_training/nnFormerTrainerV2_Synapse.py", line 339, in do_split
self.dataset_tr[i] = self.dataset[i]
KeyError: 'img0005'

282857341 · 2021-12-31T16:46:02Z

看样子应该是 self.dataset 没有img0005这个key，你可以试着把self.dataset 的key都打印出来，正常来说应该会有30个key。
运行nnFormer_plan_and_preprocess 之后会根据imagesTr和labelsTr的内容，生成nnFormer_preprocessed,
self.dateset 会从nnFormerData_plans_v2.1_stage1文件夹里所有的npz文件的名称作为key。

如果self.dataset的key少于30个，说明imagesTr和imagesTr少数据了，正常来说imagesTr和labelsTr里的数据也应该是30个

smacaijicoder · 2021-12-31T16:57:13Z

我将self.dataset打印出来了，下面是一例的内容
('AD_067', OrderedDict([('data_file', '/home/yiqingwen/nnformer/DATASET/nnFormer_preprocessed/Task560_AorticDissection/nnFormerData_plans_v2.1_stage1/AD_067.npz'), ('properties_file', '/home/yiqingwen/nnformer/DATASET/nnFormer_preprocessed/Task560_AorticDissection/nnFormerData_plans_v2.1_stage1/AD_067.pkl'), ('properties', OrderedDict([('original_size_of_raw_data', array([1374, 512, 512])), ('original_spacing', array([1., 1., 1.])), ('list_of_data_files', ['/home/yiqingwen/nnformer/DATASET/nnFormer_raw/nnFormer_raw_data/Task560_AorticDissection/imagesTr/AD_067_0000.nii.gz']), ('seg_file', '/home/yiqingwen/nnformer/DATASET/nnFormer_raw/nnFormer_raw_data/Task560_AorticDissection/labelsTr/AD_067.nii.gz'), ('itk_origin', (0.0, 0.0, 0.0)), ('itk_spacing', (1.0, 1.0, 1.0)), ('itk_direction', (1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0)), ('crop_bbox', [[0, 1374], [0, 512], [0, 512]]), ('classes', array([-1., 0., 1., 2., 3.], dtype=float32)), ('size_after_cropping', (1374, 512, 512)), ('use_nonzero_mask_for_norm', OrderedDict([(0, False)])), ('size_after_resampling', (1374, 512, 512)), ('spacing_after_resampling', array([1., 1., 1.])), ('class_locations', {1: array([[519, 276, 288],
[582, 281, 286],
[245, 275, 305],
...,
[482, 278, 283],
[179, 278, 299],
[303, 287, 274]]), 2: array([[778, 247, 262],
[382, 239, 241],
[859, 251, 248],
...,
[519, 295, 274],
[346, 225, 224],
[680, 272, 257]]), 3: array([[198, 229, 237],
[972, 249, 221],
[945, 248, 227],
...,
[186, 253, 273],
[716, 273, 291],
[164, 230, 246]])})]))]))])
病例是有68例的，报错不会是因为开头的'AD_067'和‘img0005’的key对应不上吧？如果是这样的话，我应该将311-314中的key全部改成‘AD_0XX’的格式吗？

282857341 · 2022-01-01T03:03:06Z

是的，如果你要用其他数据的话，就要替换掉

smacaijicoder · 2022-01-02T13:50:35Z

感谢！读取的error解决了，但又出现了新的问题。。。报错信息如下
Traceback (most recent call last):
File "/home/yiqingwen/anaconda3/envs/nnFormer/bin/nnFormer_train", line 33, in
sys.exit(load_entry_point('nnformer', 'console_scripts', 'nnFormer_train')())
File "/home/yiqingwen/nnformer/nnFormer/nnformer/run/run_training.py", line 181, in main
trainer.run_training()
File "/home/yiqingwen/nnformer/nnFormer/nnformer/training/network_training/nnFormerTrainerV2_Synapse.py", line 448, in run_training
ret = super().run_training()
File "/home/yiqingwen/nnformer/nnFormer/nnformer/training/network_training/nnFormerTrainer_synapse.py", line 319, in run_training
super(nnFormerTrainer_synapse, self).run_training()
File "/home/yiqingwen/nnformer/nnFormer/nnformer/training/network_training/network_trainer_synapse.py", line 481, in run_training
l = self.run_iteration(self.tr_gen, True)
File "/home/yiqingwen/nnformer/nnFormer/nnformer/training/network_training/nnFormerTrainerV2_Synapse.py", line 248, in run_iteration
output = self.network(data)
File "/home/yiqingwen/anaconda3/envs/nnFormer/lib/python3.6/site-packages/torch/nn/modules/module.py", line 889, in _call_impl
result = self.forward(*input, **kwargs)
File "/home/yiqingwen/nnformer/nnFormer/nnformer/network_architecture/Swin_Unet_l_gelunorm.py", line 928, in forward
skips = self.model_down(x)
File "/home/yiqingwen/anaconda3/envs/nnFormer/lib/python3.6/site-packages/torch/nn/modules/module.py", line 889, in _call_impl
result = self.forward(*input, **kwargs)
File "/home/yiqingwen/nnformer/nnFormer/nnformer/network_architecture/Swin_Unet_l_gelunorm.py", line 781, in forward
x_out, S, H, W, x, Ws, Wh, Ww = layer(x, Ws, Wh, Ww)
File "/home/yiqingwen/anaconda3/envs/nnFormer/lib/python3.6/site-packages/torch/nn/modules/module.py", line 889, in _call_impl
result = self.forward(*input, **kwargs)
File "/home/yiqingwen/nnformer/nnFormer/nnformer/network_architecture/Swin_Unet_l_gelunorm.py", line 431, in forward
x = blk(x, attn_mask)
File "/home/yiqingwen/anaconda3/envs/nnFormer/lib/python3.6/site-packages/torch/nn/modules/module.py", line 889, in _call_impl
result = self.forward(*input, **kwargs)
File "/home/yiqingwen/nnformer/nnFormer/nnformer/network_architecture/Swin_Unet_l_gelunorm.py", line 219, in forward
assert L == S * H * W, "input feature has wrong size"
AssertionError: input feature has wrong size

L和S H W的值如下：
L 75264
S, H, W 32 32 32
请问这是什么问题呢？

282857341 · 2022-01-02T14:29:19Z

如果你的输入的crop size和synapse一致应该就不会有这个问题
crop size可以在这里改

nnFormer/nnformer/run/default_configuration.py

Lines 49 to 69 in e20bd3b

    
           if task=='Task001_ACDC': 
        
               plans['plans_per_stage'][0]['batch_size']=4 
        
               plans['plans_per_stage'][0]['patch_size']=np.array([14,160,160]) 
        
               plans['plans_per_stage'][0]['pool_op_kernel_sizes']=[[1, 2, 2], [1, 2, 2], [2, 2, 2], [2, 2, 2]] 
        
               plans['plans_per_stage'][0]['conv_kernel_sizes']=[[3,3,3],[3,3,3],[3,3,3],[3,3,3],[3,3,3]] 
        
               pickle_file = open(plans_file,'wb') 
        
               pickle.dump(plans, pickle_file) 
        
               pickle_file.close() 
        
           # I downsample the data four times in synapse but twice (z axis) in the ACDC 
        
           # if you want to design a new way, you should reassign the value of pool_op_kernel_sizes 
        
           # 2 represents downsample and 1 for not downsample, each list in the pool_op_kernel_sizes represents the stage 
        
           # when you change the pool_op_kernel_sizes,make sure change the code in the network 
        
           # conv_kernel_sizes is not important 
        
           elif task=='Task002_Synapse': 
        
               plans['plans_per_stage'][1]['batch_size']=2 
        
               plans['plans_per_stage'][1]['patch_size']=np.array([64,128,128]) 
        
               plans['plans_per_stage'][1]['pool_op_kernel_sizes']=[[2,2,2],[2,2,2],[2,2,2],[2,2,2]] 
        
               plans['plans_per_stage'][1]['conv_kernel_sizes']=[[3,3,3],[3,3,3],[3,3,3],[3,3,3],[3,3,3]] 
        
               pickle_file = open(plans_file,'wb') 
        
               pickle.dump(plans, pickle_file) 
        
               pickle_file.close()

smacaijicoder · 2022-01-03T16:05:32Z

很抱歉再次打扰了，predict时再次遇到了问题，看了很久还是不知道应该修改哪里，所用的imagesTs和imageTr都是自己的数据集，但可以跑nnunet的代码
Traceback (most recent call last):
File "/home/yiqingwen/anaconda3/envs/nnFormer/bin/nnFormer_predict", line 33, in
sys.exit(load_entry_point('nnformer', 'console_scripts', 'nnFormer_predict')())
File "/home/yiqingwen/nnformer/nnFormer/nnformer/inference/predict_simple.py", line 229, in main
step_size=step_size, checkpoint_name=args.chk)
File "/home/yiqingwen/nnformer/nnFormer/nnformer/inference/predict.py", line 637, in predict_from_folder
segmentation_export_kwargs=segmentation_export_kwargs)
File "/home/yiqingwen/nnformer/nnFormer/nnformer/inference/predict.py", line 221, in predict_cases
mixed_precision=mixed_precision)[1][None])
File "/home/yiqingwen/nnformer/nnFormer/nnformer/training/network_training/nnFormerTrainerV2_Synapse.py", line 221, in predict_preprocessed_data_return_seg_and_softmax
mixed_precision=mixed_precision)
File "/home/yiqingwen/nnformer/nnFormer/nnformer/training/network_training/nnFormerTrainer_synapse.py", line 524, in predict_preprocessed_data_return_seg_and_softmax
mixed_precision=mixed_precision)
File "/home/yiqingwen/nnformer/nnFormer/nnformer/network_architecture/neural_network.py", line 150, in predict_3D
verbose=verbose)
File "/home/yiqingwen/nnformer/nnFormer/nnformer/network_architecture/neural_network.py", line 387, in _internal_predict_3D_3Dconv_tiled
gaussian_importance_map)[0]
File "/home/yiqingwen/nnformer/nnFormer/nnformer/network_architecture/neural_network.py", line 525, in _internal_maybe_mirror_and_pred_3D
result_torch += 1 / num_results * pred
RuntimeError: The size of tensor a (4) must match the size of tensor b (14) at non-singleton dimension 1
其次是训练时间的问题，nnFormer经我的测试比nnUnet快了将近4倍，但每个epoch还需要大约6分钟，不知道这是不是因为我没有按nnUnet作者所言从源码编译pytorch，我所使用的是单张RTX 2080

282857341 · 2022-01-04T03:35:26Z

每个epoch的iteration数目是固定的，可以通过减少epoch，或者减少每个epoch的num_batches数目来减少训练时间。此外把 self.num_val_batches_per_epoch改成50把，我之前为了复现的目的，把这个值改成了167。这样能快一点。

nnFormer/nnformer/training/network_training/network_trainer_synapse.py

Lines 98 to 99 in 80fc0e0

    
           self.num_batches_per_epoch = 250 
        
           self.num_val_batches_per_epoch = 50

nnFormer/nnformer/training/network_training/nnFormerTrainerV2_Synapse.py

Line 49 in 80fc0e0

self.max_num_epochs = 1000

然后predict出错的地方，具体的shape能打印出来吗

282857341 · 2022-01-04T03:36:59Z

顺便问一下，你的数据加上背景是几类

smacaijicoder · 2022-01-04T06:35:47Z

pred type <class 'torch.Tensor'>
pred shape torch.Size([1, 14, 64, 128, 128])
result_torch type <class 'torch.Tensor'>
result_torch shpae torch.Size([1, 4, 64, 128, 128])
我的数据加上背景正好是4类，应该是这个问题了，那我应该修改哪里呢？

282857341 · 2022-01-04T06:44:37Z

nnFormer/nnformer/network_architecture/Swin_Unet_l_gelunorm.py

Line 920 in 80fc0e0

    
           self.final.append(final_patch_expanding(embed_dim*2**i,num_classes,patch_size=patch_size))

本来这里代码设置输出通道是14，改成num_classes就行了

smacaijicoder · 2022-01-04T06:55:43Z

RuntimeError: Error(s) in loading state_dict for swintransformer:
size mismatch for final.0.up.weight: copying a param with shape torch.Size([192, 14, 2, 4, 4]) from checkpoint, the shape in current model is torch.Size([192, 4, 2, 4, 4]).
size mismatch for final.0.up.bias: copying a param with shape torch.Size([14]) from checkpoint, the shape in current model is torch.Size([4]).
size mismatch for final.1.up.weight: copying a param with shape torch.Size([384, 14, 2, 4, 4]) from checkpoint, the shape in current model is torch.Size([384, 4, 2, 4, 4]).
size mismatch for final.1.up.bias: copying a param with shape torch.Size([14]) from checkpoint, the shape in current model is torch.Size([4]).
size mismatch for final.2.up.weight: copying a param with shape torch.Size([768, 14, 2, 4, 4]) from checkpoint, the shape in current model is torch.Size([768, 4, 2, 4, 4]).
size mismatch for final.2.up.bias: copying a param with shape torch.Size([14]) from checkpoint, the shape in current model is torch.Size([4]).

新的问题出现了，应该是训练时保存模型的shape不对，这个问题修改了上面Swin_Unet_l_gelunorm.py里的代码可以解决吗，不过我应该要重新训练了

282857341 · 2022-01-04T07:00:04Z

把trainer加载预训练权重的代码注释就行了。

nnFormer/nnformer/training/network_training/nnFormerTrainerV2_Synapse.py

Lines 157 to 159 in 80fc0e0

    
           checkpoint = torch.load("../Pretrained_weight/pretrain_Synapse.model", map_location='cuda') 
        
           self.network.load_state_dict(checkpoint['state_dict']) 
        
           print('I am using the pre_train weight!!')

主要训练和推理的模型变了导致的这个问题，推理的时候会重新初始化网络，然后加载预训练权重的代码写在了初始化网络的代码里，你重新训练的话这个问题就不会出现了

smacaijicoder · 2022-01-04T07:26:33Z

那这样的话我是不是不能在自己的数据集上使用你们的预训练模型了呢？因为我重新训练时发现如果不注释上面三行代码还是会报错。或者我应该自己训一个预训练模型吗

MOMOANNIE · 2023-03-06T03:37:12Z

感谢！读取的error解决了，但又出现了新的问题。。。报错信息如下 Traceback (most recent call last): File "/home/yiqingwen/anaconda3/envs/nnFormer/bin/nnFormer_train", line 33, in sys.exit(load_entry_point('nnformer', 'console_scripts', 'nnFormer_train')()) File "/home/yiqingwen/nnformer/nnFormer/nnformer/run/run_training.py", line 181, in main trainer.run_training() File "/home/yiqingwen/nnformer/nnFormer/nnformer/training/network_training/nnFormerTrainerV2_Synapse.py", line 448, in run_training ret = super().run_training() File "/home/yiqingwen/nnformer/nnFormer/nnformer/training/network_training/nnFormerTrainer_synapse.py", line 319, in run_training super(nnFormerTrainer_synapse, self).run_training() File "/home/yiqingwen/nnformer/nnFormer/nnformer/training/network_training/network_trainer_synapse.py", line 481, in run_training l = self.run_iteration(self.tr_gen, True) File "/home/yiqingwen/nnformer/nnFormer/nnformer/training/network_training/nnFormerTrainerV2_Synapse.py", line 248, in run_iteration output = self.network(data) File "/home/yiqingwen/anaconda3/envs/nnFormer/lib/python3.6/site-packages/torch/nn/modules/module.py", line 889, in _call_impl result = self.forward(*input, **kwargs) File "/home/yiqingwen/nnformer/nnFormer/nnformer/network_architecture/Swin_Unet_l_gelunorm.py", line 928, in forward skips = self.model_down(x) File "/home/yiqingwen/anaconda3/envs/nnFormer/lib/python3.6/site-packages/torch/nn/modules/module.py", line 889, in _call_impl result = self.forward(*input, **kwargs) File "/home/yiqingwen/nnformer/nnFormer/nnformer/network_architecture/Swin_Unet_l_gelunorm.py", line 781, in forward x_out, S, H, W, x, Ws, Wh, Ww = layer(x, Ws, Wh, Ww) File "/home/yiqingwen/anaconda3/envs/nnFormer/lib/python3.6/site-packages/torch/nn/modules/module.py", line 889, in _call_impl result = self.forward(*input, **kwargs) File "/home/yiqingwen/nnformer/nnFormer/nnformer/network_architecture/Swin_Unet_l_gelunorm.py", line 431, in forward x = blk(x, attn_mask) File "/home/yiqingwen/anaconda3/envs/nnFormer/lib/python3.6/site-packages/torch/nn/modules/module.py", line 889, in _call_impl result = self.forward(*input, **kwargs) File "/home/yiqingwen/nnformer/nnFormer/nnformer/network_architecture/Swin_Unet_l_gelunorm.py", line 219, in forward assert L == S * H * W, "input feature has wrong size" AssertionError: input feature has wrong size

L和S H W的值如下： L 75264 S, H, W 32 32 32 请问这是什么问题呢？

你好，请问这个问题你是怎么解决的呢？

smacaijicoder · 2023-03-06T03:37:33Z

已收到邮件，将尽快回复！

congcongwy51 · 2024-03-27T10:09:03Z

感谢！读取的error解决了，但又出现了新的问题。。。报错信息如下 Traceback （the recent call last）：文件 “/home/yiqingwen/anaconda3/envs/nnFormer/bin/nnFormer_train”， line 33， in sys.exit（load_entry_point（'nnformer'， 'console_scripts'， 'nnFormer_train'）（））文件 “/home/yiqingwen/nnformer/nnformer/nnFormer/nnformer/run/run_training.py”，第 181 行，在主 trainer.run_training（）文件 “/home/yiqingwen/nnformer/nnFormer/nnformer/training/network_training/nnFormerTrainerV2_Synapse.py”，第 448 行，在 run_training ret = super（）.run_training（）文件“/home/yiqingwen/nnformer/nnFormer/nnformer/training/network_training/nnFormerTrainer_synapse.py”，第 319 行，在 run_training super（nnFormerTrainer_synapse， self）.run_training（）文件“/home/yiqingwen/nnformer/nnformer/nnFormer/nnformer/training/network_training/network_trainer_synapse.py”，第 481 行，在 run_training l = self.run_iteration（self.tr_gen， True）文件“/home/yiqingwen/nnformer/nnformer/nnFormer/nnformer/training/network_training/nnFormerTrainerV2_Synapse.py”，第 248 行，在run_iteration输出 = self.network（data）文件“/home/yiqingwen/anaconda3/envs/nnFormer/lib/python3.6/site-packages/torch/nn/modules/module.py”，第 889 行，在 _call_impl result = self.forward（*input， **kwargs）文件“/home/yiqingwen/nnformer/nnFormer/nnformer/network_architecture/Swin_Unet_l_gelunorm.py”，第 928 行，正向跳过 = self.model_down（x）文件“/home/yiqingwen/anaconda3/envs/nnFormer/lib/python3.6/site-packages/torch/nn/modules/module.py”，第 889 行，在 _call_impl result = self.forward（*input， **kwargs）文件“/home/yiqingwen/nnformer/nnFormer/nnformer/network_architecture/Swin_Unet_l_gelunorm.py”，第 781 行，在正向 x_out中，S、H、W、x、Ws、Wh、Ww = layer（x， Ws， Wh， Ww）文件“/home/yiqingwen/anaconda3/envs/nnFormer/lib/python3.6/site-packages/torch/nn/modules/module.py”，第 889 行，在 _call_impl result = self.forward（*input， **kwargs）文件 “/home/yiqingwen/nnformer/nnFormer/nnformer/network_architecture/Swin_Unet_l_gelunorm.py”，第 431 行，向前 x = blk（x， attn_mask）文件 “/home/yiqingwen/anaconda3/envs/nnFormer/lib/python3.6/site-packages/torch/nn/modules/module.py”，第 889 行，在 _call_impl result = self.forward（*input， **kwargs）文件 “/home/yiqingwen/nnformer/nnformer/nnFormer/nnformer/network_architecture/Swin_Unet_l_gelunorm.py”，第 219 行，在正向断言 L == S * H * W，“输入特征大小错误” AssertionError：输入特征大小错误
L和S H W的值如下： L 75264 S， H， W 32 32 32 请问这是什么问题呢？

你好，请问这个问题你是怎么解决的呢？

请问您解决了嘛

smacaijicoder · 2024-03-27T10:09:32Z

已收到邮件，将尽快回复！

congcongwy51 · 2024-03-27T10:25:27Z

splits[self.fold]['train']=np.array(['img0006','img0007' ,'img0009', 'img0010', 'img0021' ,'img0023' ,'img0024','img0026' ,'img0027' ,'img0031', 'img0033' ,'img0034'
,'img0039', 'img0040','img0005', 'img0028', 'img0030', 'img0037'])
splits[self.fold]['val']=np.array(['img0001', 'img0002', 'img0003', 'img0004', 'img0008', 'img0022','img0025', 'img0029', 'img0032', 'img0035', 'img0036', 'img0038']) 我想问问这段代码放的数据是将我得训练集分为train和val吗？不需要加测试集是吗？

congcongwy51 · 2024-03-27T13:44:24Z

如果你的输入的crop size和synapse一致应该就不会有这个问题 crop size可以在这里改

nnFormer/nnformer/run/default_configuration.py

Lines 49 to 69 in e20bd3b

if task=='Task001_ACDC':

plans['plans_per_stage'][0]['batch_size']=4

plans['plans_per_stage'][0]['patch_size']=np.array([14,160,160])

plans['plans_per_stage'][0]['pool_op_kernel_sizes']=[[1, 2, 2], [1, 2, 2], [2, 2, 2], [2, 2, 2]]

plans['plans_per_stage'][0]['conv_kernel_sizes']=[[3,3,3],[3,3,3],[3,3,3],[3,3,3],[3,3,3]]

pickle_file = open(plans_file,'wb')

pickle.dump(plans, pickle_file)

pickle_file.close()

# I downsample the data four times in synapse but twice (z axis) in the ACDC

# if you want to design a new way, you should reassign the value of pool_op_kernel_sizes

# 2 represents downsample and 1 for not downsample, each list in the pool_op_kernel_sizes represents the stage

# when you change the pool_op_kernel_sizes,make sure change the code in the network

# conv_kernel_sizes is not important

elif task=='Task002_Synapse':

plans['plans_per_stage'][1]['batch_size']=2

plans['plans_per_stage'][1]['patch_size']=np.array([64,128,128])

plans['plans_per_stage'][1]['pool_op_kernel_sizes']=[[2,2,2],[2,2,2],[2,2,2],[2,2,2]]

plans['plans_per_stage'][1]['conv_kernel_sizes']=[[3,3,3],[3,3,3],[3,3,3],[3,3,3],[3,3,3]]

pickle_file = open(plans_file,'wb')

pickle.dump(plans, pickle_file)

pickle_file.close()

我是S， H， W 32 32 32，我应该怎么修改呀

Jx-Tan mentioned this issue Jan 11, 2022

the means of self.num_batches_per_epoch and self.num_val_batches_per_epoch？ #48

Open

MOMOANNIE mentioned this issue Mar 6, 2023

训练自己的数据，出现尺寸不对的问题 #110

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

train error #44

train error #44

smacaijicoder commented Dec 31, 2021

282857341 commented Dec 31, 2021

smacaijicoder commented Dec 31, 2021 via email

282857341 commented Dec 31, 2021

smacaijicoder commented Dec 31, 2021 via email

282857341 commented Dec 31, 2021

smacaijicoder commented Dec 31, 2021 via email

smacaijicoder commented Dec 31, 2021

282857341 commented Dec 31, 2021

smacaijicoder commented Dec 31, 2021 •

edited

Loading

282857341 commented Jan 1, 2022

smacaijicoder commented Jan 2, 2022

282857341 commented Jan 2, 2022

smacaijicoder commented Jan 3, 2022

282857341 commented Jan 4, 2022

282857341 commented Jan 4, 2022

smacaijicoder commented Jan 4, 2022

282857341 commented Jan 4, 2022

smacaijicoder commented Jan 4, 2022

282857341 commented Jan 4, 2022

smacaijicoder commented Jan 4, 2022

MOMOANNIE commented Mar 6, 2023

smacaijicoder commented Mar 6, 2023 via email

congcongwy51 commented Mar 27, 2024

smacaijicoder commented Mar 27, 2024 via email

congcongwy51 commented Mar 27, 2024

congcongwy51 commented Mar 27, 2024

train error #44

train error #44

Comments

smacaijicoder commented Dec 31, 2021

282857341 commented Dec 31, 2021

smacaijicoder commented Dec 31, 2021 via email

282857341 commented Dec 31, 2021

smacaijicoder commented Dec 31, 2021 via email

282857341 commented Dec 31, 2021

smacaijicoder commented Dec 31, 2021 via email

smacaijicoder commented Dec 31, 2021

282857341 commented Dec 31, 2021

smacaijicoder commented Dec 31, 2021 • edited Loading

282857341 commented Jan 1, 2022

smacaijicoder commented Jan 2, 2022

282857341 commented Jan 2, 2022

smacaijicoder commented Jan 3, 2022

282857341 commented Jan 4, 2022

282857341 commented Jan 4, 2022

smacaijicoder commented Jan 4, 2022

282857341 commented Jan 4, 2022

smacaijicoder commented Jan 4, 2022

282857341 commented Jan 4, 2022

smacaijicoder commented Jan 4, 2022

MOMOANNIE commented Mar 6, 2023

smacaijicoder commented Mar 6, 2023 via email

congcongwy51 commented Mar 27, 2024

smacaijicoder commented Mar 27, 2024 via email

congcongwy51 commented Mar 27, 2024

congcongwy51 commented Mar 27, 2024

smacaijicoder commented Dec 31, 2021 •

edited

Loading