Pytorch implementation of U-Net, R2U-Net, Attention U-Net, and Attention R2U-Net.

hello i have some problems about this code Traceback (most recent call last): File "/home/panpan/wenjian/pytorchImage_Segmentation-master/dataset.py", line 115, in main(config) File "/home/panpan/wenjian/pytorchImage_Segmentation-master/dataset.py", line 55, in main copyfile(src, dst) File "/home/panpan/anaconda3/lib/python3.6/shutil.py", line 120, in copyfile with open(src, 'rb') as fsrc: FileNotFoundError: [Errno 2] No such file or directory: '../ISIC/dataset/train_data/ISIC_33.png'

i dont know how to deal with ,can you help me ?thanks a lot

hello ,thank you very much for your contribution

could you tell me how to test the code ,because

Epoch [15/15], Loss: 11.5097, [Training] Acc: 0.9289, SE: 0.8480, SP: 0.9742, PC: 0.8885, F1: 0.8537, JS: 0.7518, DC: 0.8537 Decay learning rate to lr: 0.0. [Validation] Acc: 0.8809, SE: 0.9680, SP: 0.8651, PC: 0.6196, F1: 0.7209, JS: 0.5988, DC: 0.7209

Process finished with exit code 0

the test function ,doesn't work , and i only can see the test result

Hello, @LeeJunHyun I followed your instruction by changing the path to my data in dataset.py L99,100 .. but after running main.py i got errors in L152(solver.py) about the unequal size of SR_Flat, GT_Flat !!

ValueError: Target and input must have the same number of elements. target nelement (196608) != input nelement (65536)

Jaccard and F1 only 0.5-0.7,are you sure? On the official competition website, the result common is 0.80+ in term of jaccard. In addition, the result of Unet in my experiment is 0.78+ in term of jaccard, Yes! ISIC 2018. 2594 images.I resize it to 256*192

Hey Lee, So I am wanted to implement Attention Unet. But there is some issue coming up while using it. I am using my own dataset and also doing a lot of preprocessing with data. When I am using your model I get this error.

Any idea about it? Its for medical binary segmentation

I have 2 questions. 1.What kind of images can I use? My input includes single channel '.png' images which should be segmented and '.png' GT images which is composed of 0 and 255. Can I use them directly, or what is your image format?

2. What does it mean that SR = SR > treshhold in get_accuracy in evaluation? For example, my SR.max() is 8 , SR.min() is -10 for a instance. Are you supposed SR is ranged in (0:1)? Why did I get this wried results?

Thank you so much

[Training] Acc: 0.6241, SE: 0.0000, SP: 0.0000, PC: 0.0000, F1: 0.0000, JS: 0.0000, DC: 0.0000 [Validation] Acc: 0.7816, SE: 0.0000, SP: 0.0000, PC: 0.0000, F1: 0.0000, JS: 0.0000, DC: 0.0000 Epoch [42/250], Loss: 0.0649, [Training] Acc: 0.6238, SE: 0.0000, SP: 0.0000, PC: 0.0000, F1: 0.0000, JS: 0.0000, DC: 0.0000 [Validation] Acc: 0.7816, SE: 0.0000, SP: 0.0000, PC: 0.0000, F1: 0.0000, JS: 0.0000, DC: 0.0000 Epoch [43/250], Loss: 0.0613, [Training] Acc: 0.6243, SE: 0.0000, SP: 0.0000, PC: 0.0000, F1: 0.0000, JS: 0.0000, DC: 0.0000 [Validation] Acc: 0.7814, SE: 0.0000, SP: 0.0000, PC: 0.0000, F1: 0.0000, JS: 0.0000, DC: 0.0000 Epoch [44/250], Loss: 0.0590, [Training] Acc: 0.6240, SE: 0.0000, SP: 0.0000, PC: 0.0000, F1: 0.0000, JS: 0.0000, DC: 0.0000 [Validation] Acc: 0.7815, SE: 0.0000, SP: 0.0000, PC: 0.0000, F1: 0.0000, JS: 0.0000, DC: 0.0000 Epoch [45/250], Loss: 0.0560, [Training] Acc: 0.6242, SE: 0.0000, SP: 0.0000, PC: 0.0000, F1: 0.0000, JS: 0.0000, DC: 0.0000 [Validation] Acc: 0.7815, SE: 0.0000, SP: 0.0000, PC: 0.0000, F1: 0.0000, JS: 0.0000, DC: 0.0000 Epoch [46/250], Loss: 0.0539, [Training] Acc: 0.6240, SE: 0.0000, SP: 0.0000, PC: 0.0000, F1: 0.0000, JS: 0.0000, DC: 0.0000 [Validation] Acc: 0.7815, SE: 0.0000, SP: 0.0000, PC: 0.0000, F1: 0.0000, JS: 0.0000, DC: 0.0000 Epoch [47/250], Loss: 0.0517, [Training] Acc: 0.6239, SE: 0.0000, SP: 0.0000, PC: 0.0000, F1: 0.0000, JS: 0.0000, DC: 0.0000 [Validation] Acc: 0.7816, SE: 0.0000, SP: 0.0000, PC: 0.0000, F1: 0.0000, JS: 0.0000, DC: 0.0000 Epoch [48/250], Loss: 0.0492, [Training] Acc: 0.6242, SE: 0.0000, SP: 0.0000, PC: 0.0000, F1: 0.0000, JS: 0.0000, DC: 0.0000 PS C:\Users\Egp\Desktop\Image_Segmentation-master>

i train on my own data. i don'know why. can u help me ?

Hi: thank you for the code, i have trained the unet with the isbi dataset 100 epoch,batch size is 8(30 gray picture),but the result is very bad, about 11 percent(accuracy). And i wonder is there something wrong that i have done, could you give me some advise? data_loader .txt main .txt I just modify the data_loader.py and change the input channel to 1(the picture is gray)

Namespace(augmentation_prob=0.16924156139739033, batch_size=4, beta1=0.5, beta2=0.999, cuda_idx=3, image_size=224, img_ch=1, log_step=2, lr=0.00017304986946574825, mode='train', model_path='./models', model_type='R2U_Net', num_epochs=250, num_epochs_decay=186, num_workers=8, output_ch=1, result_path='./result/R2U_Net', t=3, test_path='/home/Data/DC_disk2/tsachi_dataset/dataset/test/', train_path='/home/Data/DC_disk2/tsachi_dataset/dataset/train/', val_step=2, valid_path='/home/Data/DC_disk2/tsachi_dataset/dataset/valid/') image count in train path :2400 image count in valid path :211 image count in test path :200 Traceback (most recent call last): File "main.py", line 101, in main(config) File "main.py", line 61, in main solver.train() File "/home/imagry/tsachi/solver.py", line 140, in train for i, (images, GT) in enumerate(self.train_loader): File "/usr/local/lib/python3.5/dist-packages/torch/utils/data/dataloader.py", line 336, in next return self._process_next_batch(batch) File "/usr/local/lib/python3.5/dist-packages/torch/utils/data/dataloader.py", line 357, in _process_next_batch raise batch.exc_type(batch.exc_msg) RuntimeError: Traceback (most recent call last): File "/usr/local/lib/python3.5/dist-packages/torch/utils/data/dataloader.py", line 106, in _worker_loop samples = collate_fn([dataset[i] for i in batch_indices]) File "/usr/local/lib/python3.5/dist-packages/torch/utils/data/dataloader.py", line 187, in default_collate return [default_collate(samples) for samples in transposed] File "/usr/local/lib/python3.5/dist-packages/torch/utils/data/dataloader.py", line 187, in return [default_collate(samples) for samples in transposed] File "/usr/local/lib/python3.5/dist-packages/torch/utils/data/dataloader.py", line 164, in default_collate return torch.stack(batch, 0, out=out) RuntimeError: invalid argument 0: Sizes of tensors must match except in dimension 0. Got 192 and 320 in dimension 2 at /pytorch/aten/src/TH/generic/THTensorMath.cpp:3616

Exception ignored in: <bound method _DataLoaderIter.del of <torch.utils.data.dataloader._DataLoaderIter object at 0x7fca6e95b7b8>> Traceback (most recent call last): File "/usr/local/lib/python3.5/dist-packages/torch/utils/data/dataloader.py", line 399, in del self._shutdown_workers() File "/usr/local/lib/python3.5/dist-packages/torch/utils/data/dataloader.py", line 378, in _shutdown_workers self.worker_result_queue.get() File "/usr/lib/python3.5/multiprocessing/queues.py", line 345, in get return ForkingPickler.loads(res) File "/usr/local/lib/python3.5/dist-packages/torch/multiprocessing/reductions.py", line 151, in rebuild_storage_fd fd = df.detach() File "/usr/lib/python3.5/multiprocessing/resource_sharer.py", line 57, in detach with _resource_sharer.get_connection(self._id) as conn: File "/usr/lib/python3.5/multiprocessing/resource_sharer.py", line 87, in get_connection c = Client(address, authkey=process.current_process().authkey) File "/usr/lib/python3.5/multiprocessing/connection.py", line 487, in Client c = SocketClient(address) File "/usr/lib/python3.5/multiprocessing/connection.py", line 614, in SocketClient s.connect(address) ConnectionRefusedError: [Errno 111] Connection refused

I change the background to 0 and others to 1 in my labels.But when I save the GT that is loaded from dataloader as a image, I get a image full of 0. Do you know why it happened?

First of all, thank you for your code where I do learned a lot from. I work on a dataset looks like isic2018 where only got one category and around 900 images for training.

But I occur to a problem that when training, JS would keep equal to 1 and DC would even keeps greater than 1. I can't find reason since normalization in data_loader has already make sure DC could not larger than 1. Do you have any ideas?

I noticed things would happens when I use R2Unet and when SR output is totally black(nothing is divided out), maybe that would be helpful to find out where got something wrong.

Another problem is it seems the model cannot learn things during epoches processing, the best model would come out in first 5 epoches when training for like 200 epoches. Is that because model would get better result on small set and when it generalized to more images the score would drop? By the way, implementing only U-net would not get DC>1 and get some result.

Thank you in advance for your help~

Edit: Just found out JS would also stuck on 1.0 and DC greater than 1 when training Attention U-net but this time it get some result. So it maybe just JS and DC calculation problem No way this time same situation happens to U-net and I changed nothing...

Thank you for sharing your amazing code;

the "length" variable in the solver is the total data size, how in the evaluation.py ,the acc and othor metrics is calculated base on the batch, so the lengeth variable should be the iterations (base on batch), not the total number of the train data

hey guys. use pytorch<=1.2.0 (not confirmed) or change the funtion in "evaluation" to fit the calculation of bool tensor will sovle the problem e.g. change "TP = ((SR==1)+(GT==1))==2" to "TP = SR & GT"

thanks, i try it now, and do you have some recommend with pytorch > 1.2?

Originally posted by @CodeXiaoLingYun in https://github.com/LeeJunHyun/Image_Segmentation/issues/58#issuecomment-741368638

Traceback (most recent call last): File "/home/pxg/alex/U-Net/main.py", line 101, in main(config) File "/home/pxg/alex/U-Net/main.py", line 61, in main solver.train() File "/home/pxg/alex/U-Net/solver.py", line 140, in train for i, (images, GT) in enumerate(self.train_loader): File "/home/pxg/DANet-master/venv/lib/python3.6/site-packages/torch/utils/data/dataloader.py", line 819, in next return self._process_data(data) File "/home/pxg/DANet-master/venv/lib/python3.6/site-packages/torch/utils/data/dataloader.py", line 846, in _process_data data.reraise() File "/home/pxg/DANet-master/venv/lib/python3.6/site-packages/torch/_utils.py", line 369, in reraise raise self.exc_type(msg) RuntimeError: Caught RuntimeError in DataLoader worker process 0. Original Traceback (most recent call last): File "/home/pxg/DANet-master/venv/lib/python3.6/site-packages/torch/utils/data/_utils/worker.py", line 178, in _worker_loop data = fetcher.fetch(index) File "/home/pxg/DANet-master/venv/lib/python3.6/site-packages/torch/utils/data/_utils/fetch.py", line 47, in fetch return self.collate_fn(data) File "/home/pxg/DANet-master/venv/lib/python3.6/site-packages/torch/utils/data/_utils/collate.py", line 80, in default_collate return [default_collate(samples) for samples in transposed] File "/home/pxg/DANet-master/venv/lib/python3.6/site-packages/torch/utils/data/_utils/collate.py", line 80, in return [default_collate(samples) for samples in transposed] File "/home/pxg/DANet-master/venv/lib/python3.6/site-packages/torch/utils/data/_utils/collate.py", line 56, in default_collate return torch.stack(batch, 0, out=out) RuntimeError: invalid argument 0: Sizes of tensors must match except in dimension 0. Got 160 and 384 in dimension 2 at /pytorch/aten/src/TH/generic/THTensor.cpp:689

Hello, can you help me see what the problem is? This is the result of batch_size = 8. When batch_size = 1, the computer gets stuck, can't move at all, and can only be restarted. Do you know what's going on? I hope you're free to help. Thank you.

Hello, SORRY to bother you, when I refocused m attention Unet on my custom dataset, I also encountered JS which was always 1 and DICE >1, I saw previous comments, I changed Torch. Max to Max without any improvement, and my loss was always over 1000, what is the reason?