Hi there! Thank you for your innovative work. I want to ask how your Unet and attention-unet are trained. I tried many times and I couldn’t achieve that effect.

Looking forward to your reply. Best wish!!!

Hi, thanks for your contribution. The model is unable to adapt to different resolutions except for 224 for both height and width. The error often occurs as

RuntimeError: shape '[xxx, xxx, xxx, xxx, xxx]' is invalid for input of size xxx

hello @HuCaoFighting , I want to use the swin_unet for segmentation task, and the data is end with png or jpg rather than .nii.gz. So I need to change the dataset_synapse.py, I tried many times but I failed. Coul you give me some advice? Thanks!

Hi! Thanks for your nice work! And I'd like to ask what is the preprocessing method of the ACDC dataset? I did not find relevant information in TransUNet and Swin-Unet.

When I train with my datasets, the error occured as below:

=> merge config from configs/swin_tiny_patch4_window7_224_lite.yaml
SwinTransformerSys expand initial----depths:[2, 2, 2, 2];depths_decoder:[1, 2, 2, 2];drop_path_rate:0.2;num_classes:9
---final upsample expand_first---
pretrained_path:./pretrained_ckpt/swin_tiny_patch4_window7_224.pth
Traceback (most recent call last):
  File "train.py", line 96, in <module>
    net.load_from(config)
  File "/mnt/e/projects/Sementic_Segmentation/Swin-Unet-PyTorch/networks/vision_transformer.py", line 58, in load_from
    pretrained_dict = torch.load(pretrained_path, map_location=device)
  File "/home/byronnar/anaconda3/envs/swin_unet_torch/lib/python3.7/site-packages/torch/serialization.py", line 527, in load
    with _open_zipfile_reader(f) as opened_zipfile:
  File "/home/byronnar/anaconda3/envs/swin_unet_torch/lib/python3.7/site-packages/torch/serialization.py", line 224, in __init__
    super(_open_zipfile_reader, self).__init__(torch._C.PyTorchFileReader(name_or_buffer))
RuntimeError: version_ <= kMaxSupportedFileFormatVersion INTERNAL ASSERT FAILED at /pytorch/caffe2/serialize/inline_container.cc:132, please report a bug to PyTorch. Attempted to read a PyTorch file with version 3, but the maximum supported version for reading is 2. Your PyTorch installation may be too old. (init at /pytorch/caffe2/serialize/inline_container.cc:132)
frame #0: c10::Error::Error(c10::SourceLocation, std::string const&) + 0x33 (0x7fd2bbfab193 in /home/byronnar/anaconda3/envs/swin_unet_torch/lib/python3.7/site-packages/torch/lib/libc10.so)
frame #1: caffe2::serialize::PyTorchStreamReader::init() + 0x1f5b (0x7fd2bf1339eb in /home/byronnar/anaconda3/envs/swin_unet_torch/lib/python3.7/site-packages/torch/lib/libtorch.so)
frame #2: caffe2::serialize::PyTorchStreamReader::PyTorchStreamReader(std::string const&) + 0x64 (0x7fd2bf134c04 in /home/byronnar/anaconda3/envs/swin_unet_torch/lib/python3.7/site-packages/torch/lib/libtorch.so)
frame #3: <unknown function> + 0x6c6536 (0x7fd307246536 in /home/byronnar/anaconda3/envs/swin_unet_torch/lib/python3.7/site-packages/torch/lib/libtorch_python.so)
frame #4: <unknown function> + 0x295a74 (0x7fd306e15a74 in /home/byronnar/anaconda3/envs/swin_unet_torch/lib/python3.7/site-packages/torch/lib/libtorch_python.so)
<omitting python frames>
frame #33: __libc_start_main + 0xf3 (0x7fd319cf10b3 in /lib/x86_64-linux-gnu/libc.so.6)

How should I do? Thank you!

Traceback (most recent call last): File "train.py", line 10, in from config import get_config File "C:\Users\Downloads\Swin-Unet-main\config.py", line 72, in _C.MODEL.SWIN.QK_SCALE = None File "C:\anaconda\envs\swin\lib\site-packages\yacs\config.py", line 158, in setattr type(value), name, _VALID_TYPES File "C:\anaconda\envs\swin\lib\site-packages\yacs\config.py", line 521, in _assert_with_logging assert cond, msg AssertionError: Invalid type <class 'NoneType'> for key QK_SCALE; valid types = {<class 'float'>, <class 'tuple'>, <class 'str'>, <class 'list'>, <class 'bool'>, <class 'int'>}

I'm trying to reproduce the experiment. The training stage looks good. However, the test stage gets stuck here around half an hour. Is this normal? I also wonder how long it will take to finish the evaluation? Thanks!

Dear researchers, please also consider checking our new unet implementation of the proposed focal modulation block for small datasets problems, specially in medical domains. https://github.com/givkashi/Focal-Unet

https://www.researchgate.net/publication/366423296_Focal-UNet_UNet-like_Focal_Modulation_for_Medical_Image_Segmentation

Thanks.

I want to translate this network with torchsharp
in class PatchExpand 语言用C#

   x = x.view(B, H, W, C);
   x1 = x.reshape(B,H*2,W*2,C/4);

   x2= rearrange(x, 'b h w (p1 p2 c)-> b (h p1) (w p2) c', p1 : 2, p2 : 2, c : C / 4)

Can x1 replace x2? Or there are other options？

First I would like to thank you for your contribution. I was wondering, how many GPUs did you use and how long does it take to train your network and reach the results of the paper? Thanks in advance for the answer.

During training, the parameters I use are: accumulation_steps=None, amp_opt_level='O1', base_lr=0.01, batch_size=6, cache_mode='part', cfg='./configs/swin_tiny_patch4_window7_224_lite.yaml', dataset='Synapse', deterministic=1, eval=False, img_size=224, is_pretrain=True, is_savenii=False, list_dir='./lists/lists_Synapse', max_epochs=150, max_iterations=30000, num_classes=14, opts=None, output_dir='./output_dir', resume=None, seed=1234, tag=None, test_save_dir='../predictions', throughput=False, use_checkpoint=False, volume_path='../data/Synapse/test_vol_h5', z_spacing=1, zip=False) epoch_149.pth The difference with method in paper is: num_classes=14, because there 13 classes and background in Synapse dataset. However, I got <mean_dice : 0.000471 mean_hd95 : 56.415742 I don't know the reason. Can somebody help me?