Hello @junqiangchen,
Thank you for your project
I have tried to predict some data-set but results are not good.
Maybe source image which is converted from *.nii file wrong.
I use https://github.com/kirsebaer/nii2img for convert *.nii to *.bmp
Could you please sharing code to convert .nii to .bmp image?

Hi junqiangchen,

Can you provide a lead over the conversion of prediction results to .mp4 format?

I notice that the default GPU setting in this training code is :
os.environ["CUDA_VISIBLE_DEVICES"] = "0" and batch_size = 1

when I simply change them to:
os.environ["CUDA_VISIBLE_DEVICES"] = "0,1,2,3,4,5,6,7" and batch_size = 8
(since I have 8 GPUs)
OOM error happens

so can this code support GPU parallel naturally or should I modify it for mutli-GPU scenario?
my tf version is 1.8.0

Thank you!

Hello, thanks for making your code public, and your code is great, I have run your code, but i couldn't test my data on the LITS website, Can you tell me how to do that? Thank you.

I'm a little confused about which code I want to run

I got this error. "UnicodeEncodeError: 'utf-8' codec can't encode character '\udcfd' in position 3391: surrogates not allowed"

Dear,
with the resources, I can't download because it all Chinese. I try to register but not allow any thai phone number. Could you help me?

liver segmentation trained model,log,test data can download on here:https://pan.baidu.com/s/1ijK6BG3vZM4nHwZ6S2yFiw, password：74j5
LiTS data have 130 cases,using 0-110 cases trainging,and other is testing.testing result can download on here:https://pan.baidu.com/s/1A_-u7tJcn7rIqnrLaSqi4A password：22es
LiTS train and test source data can download here:https://pan.baidu.com/s/1-kxJ7reS4kq5ypitfmQxeg password：nlrd

你好，我是一个刚入门，请问运行utils.py时出现了TypeError: object of type 'NoneType' has no len()这个错误是怎么回事？

Hi,
I have problem in train file that generates some error related to cuda. can you please tell which cuda version you have used?
Thanks

It's convient to visualize the segment result with a video, but my video is some big, so i want to konw how to generate '.mp4' file like yours? I'm using 'cv2'.

Thanks for your reply~

Hello, I want to segmentation the liver and tumor at the same time, how to change it, thank you, look forward to your answer

My GPU is RTX 2070 8G, when I run vnet3d_train with patch(256,256,16), it fails, error log liike that:
Caused by op 'output/Conv3D', defined at:
File "D:/py/LiTS---Liver-Tumor-Segmentation-Challenge-master/LiTS/vnet3d_train.py", line 42, in
train()
File "D:/py/LiTS---Liver-Tumor-Segmentation-Challenge-master/LiTS/vnet3d_train.py", line 36, in train
Vnet3d = Vnet3dModule(256, 256, 16, channels=1, costname=("dice coefficient",))
File "D:\py\LiTS---Liver-Tumor-Segmentation-Challenge-master\LiTS\Vnet\model_vnet3d.py", line 203, in init
self.phase, self.drop)
File "D:\py\LiTS---Liver-Tumor-Segmentation-Challenge-master\LiTS\Vnet\model_vnet3d.py", line 152, in _create_conv_net
output_map = conv_sigmod(x=layer9, kernal=(1, 1, 1, 32, n_class), scope='output')
File "D:\py\LiTS---Liver-Tumor-Segmentation-Challenge-master\LiTS\Vnet\model_vnet3d.py", line 52, in conv_sigmod
conv = conv3d(x, W) + B
File "D:\py\LiTS---Liver-Tumor-Segmentation-Challenge-master\LiTS\Vnet\layer.py", line 51, in conv3d
conv_3d = tf.nn.conv3d(x, W, strides=[1, stride, stride, stride, 1], padding='SAME')
File "C:\Users\aiage\Miniconda3\envs\vnet_tensor\lib\site-packages\tensorflow\python\ops\gen_nn_ops.py", line 1487, in conv3d
name=name)
File "C:\Users\aiage\Miniconda3\envs\vnet_tensor\lib\site-packages\tensorflow\python\framework\op_def_library.py", line 787, in _apply_op_helper
op_def=op_def)
File "C:\Users\aiage\Miniconda3\envs\vnet_tensor\lib\site-packages\tensorflow\python\framework\ops.py", line 3392, in create_op
op_def=op_def)
File "C:\Users\aiage\Miniconda3\envs\vnet_tensor\lib\site-packages\tensorflow\python\framework\ops.py", line 1718, in init
self._traceback = self._graph._extract_stack() # pylint: disable=protected-access

InternalError (see above for traceback): Blas SGEMM launch failed : m=1048576, n=1, k=32
[[Node: output/Conv3D = Conv3D[T=DT_FLOAT, data_format="NDHWC", dilations=[1, 1, 1, 1, 1], padding="SAME", strides=[1, 1, 1, 1, 1], _device="/job:localhost/replica:0/task:0/device:GPU:0"](add_8, outputW/read/_139)]]
[[Node: Neg/_143 = _Recvclient_terminated=false, recv_device="/job:localhost/replica:0/task:0/device:CPU:0", send_device="/job:localhost/replica:0/task:0/device:GPU:0", send_device_incarnation=1, tensor_name="edge_2696_Neg", tensor_type=DT_FLOAT, _device="/job:localhost/replica:0/task:0/device:CPU:0"]]

And what can I do with it ?

您好，请问您可以提供.nii格式的原始测试数据吗？我在LITS挑战赛的官方网站上不能打开测试数据链接。

万分感谢!

my segmentation resullts of Liver is all white image,why?

Hello! I've found a performance issue in dataset_input.py: dataset.batch(batch_size)(line 109) should be called before dataset.map(_parse_function)(line 107) and dataset.map(_preprocess_function)(line 108), which could make your program more efficient.

Here is the tensorflow document to support it.

Besides, you need to check the function _preprocess_function called in dataset.map(_preprocess_function) whether to be affected or not to make the changed code work properly. For example, if _preprocess_function needs data with shape (x, y, z) as its input before fix, it would require data with shape (batch_size, x, y, z) after fix.

Looking forward to your reply. Btw, I am very glad to create a PR to fix it if you are too busy.

Hello,I found a performance issue in the definition of datagenerator ,
LiTS/data_input/dataset_input.py,
dataset = dataset.map(_parse_function) was called without num_parallel_calls.
I think it will increase the efficiency of your program if you add this.

The same issues also exist in dataset = dataset.map(_preprocess_function)

Here is the documemtation of tensorflow to support this thing.

Looking forward to your reply. Btw, I am very glad to create a PR to fix it if you are too busy.

您好，这段nii转bmp的代码是我模仿生成训练数据写的，测试的效果很差，想知道问题是出在了哪里 ：）

def prepareTestData():
    for i in range(0, 70, 1):
        filepath = testImage + str(i) + ".nii"
        src = load_itk(filepath)
        data = sitk.GetArrayFromImage(src)
        data = data.astype(np.float32)
        data = np.clip(data, 0, 255).astype('uint8')
        channel, width, height = data.shape[0], data.shape[1], data.shape[2]
        for j in range(channel):
            savePath = outputPath + str(i)
            if not os.path.exists(savePath):
                os.makedirs(savePath)
            v_flip = cv2.flip(data[j, :, :], 0)  # 上下翻转图片
            cv2.imwrite(savePath + "/" + str(j) + ".bmp", v_flip)

Here is my code:
from model_vnet3d import Vnet3dModule
from util import convertMetaModelToPbModel
import numpy as np
import pandas as pd
import cv2
import os
os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
os.environ["CUDA_VISIBLE_DEVICES"] = "0"

def train():
'''
Preprocessing for dataset
'''
# Read data set (Train data from CSV file)
csvmaskdata = pd.read_csv('E:/LiTS---Liver/LiTS/dataprocess/trainY.csv')
csvimagedata = pd.read_csv('E:/LiTS---Liver/LiTS/dataprocess/trainX.csv')
maskdata = csvmaskdata.iloc[:, :].values
imagedata = csvimagedata.iloc[:, :].values
# shuffle imagedata and maskdata together
perm = np.arange(len(csvimagedata))
np.random.shuffle(perm)
imagedata = imagedata[perm]
maskdata = maskdata[perm]

Vnet3d = Vnet3dModule(128, 128, 16, channels=1, costname="dice coefficient")
Vnet3d.train(imagedata, maskdata, "Vnet3dModule.pd", "E:/LiTS---Liver/LiTS/log2/diceVnet3d/model/", 0.001, 0.7, 100000, 1)

def predict0():
Vnet3d = Vnet3dModule(128, 128, 16, inference=True, model_path="E:\LiTS---Liver\LiTS\log2\diceVnet3d\model\Vnet3d.pd")
for filenumber in range(30):
batch_xs = np.zeros(shape=(16, 128, 128))
for index in range(16):
imgs = cv2.imread(
"E:\LiTS---Liver\LiTS\test\image\" + str(filenumber) + "\" + str(index) + ".bmp", 0)
batch_xs[index, :, :] = imgs[128:384, 128:384]

    predictvalue = Vnet3d.prediction(batch_xs)

    for index in range(16):
        result = np.zeros(shape=(512, 512), dtype=np.uint8)
        result[128:384, 128:384] = predictvalue[index]
        kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (5, 5))
        result = cv2.morphologyEx(result, cv2.MORPH_CLOSE, kernel)
        cv2.imwrite(
            "E:\LiTS---Liver\LiTS\\test\image\\" + str(filenumber) + "\\" + str(index) + "mask.bmp",
            result)

def meta2pd():
convertMetaModelToPbModel(meta_model="E:\LiTS---Liver\LiTS\log2\diceVnet3d\model\Vnet3d.pd", pb_model="model")

#train()
predict0()
#meta2pd()

my getpatchimageandmask.py always abortion

hello,
yesterday i made a submission on codalab but after running a couple of hours i was surprised by a weird ''time limit exceeded" error do you have any idea about the origin of this error ? thank you

How long does it take to run getPatchImageAndMask.py?

hello,thanks for your contribution.I want ask one question,as follow:
how do you sellect the number of sample which is in the getPatchImageAndMask.py.
I print the number of sample,which is 5917.I use the dataset of LITS.
Looking forward to hearing from you!

Hi I don't have baidu account. So, I am unable to download your model file. Can you please share any other links for this?

请教一下，在函数gen_image_mask中，您保存的是16张256×256的图片，我是否可以将这16幅图保存为一个nii文件呢？具体应该怎么操作？谢谢