iwuqing/arssr

[IEEE J-BHI] An Arbitrary Scale Super-Resolution Approach for 3D MR Images using Implicit Neural Representation

arssr's Issues

Which HCP-1200 data set of skull dissection was used in the paper?

I am running HCP datasets for the first time and do not know which is the dataset described in this article.

some questions

你好！
关于ArSSR有些问题想问下你，麻烦了！
问题1：关于ArSSR的LPC-SI的无参考评价指标，我下载了其matlab的代码，输入归一化后的切片，其得分在0.07xx，非常低。但是如果输入没有归一化的切片，原始MR 图像，其得分在0.99xx,得分很正常，这里我很困惑，想问下你是怎么操作的？
问题2：关于论文的分割评价（Fully Automatic Segmentation based Evaluation）这部的代码实现或者使用教程能否提供让学习一下。
问题3：关于图像的归一化处理，用的SimpleITK自带方法？还是 简单的（x - min）/ （max - min）进行（0,1）的归一化？下载的原始MR图像的体素最大值有的是1400多，也有2000多的，这里归一化的时候我也是很困惑。
​resacleFilter = sitk.RescaleIntensityImageFilter()
resacleFilter.SetOutputMaximum(1)
resacleFilter.SetOutputMinimum(0)
image = resacleFilter.Execute(image)
很期待你的回复！祝你科研顺利，生活愉快！

能否提供一下测试数据集

你好，请问方便提供一下HCP-1200预处理过的测试集吗

HCP_1200 t1w 原始数据

作者您好，我用您发的脚本下载的HCP_1200 t1w的图像数据，尺寸大小为（260，311，260），并不是您论文里尺寸为（320320256）您可以再提供一下HCP-1200 T1W的原始数据的网盘链接吗（之前的过期了）。第二个问题是，您是如何讲图片裁剪到264264264的？万分感谢您可以抽空回答我。邮箱：[email protected]

评价指标的代码

Hello，首先很感谢您和您的团队把训练和测试相关工作开源，能否把训练数据集的预处理和评价指标的相关代码也分享，供大家学习！

Wrong command in README

I used the command mentioned in the README to test the pre-trained model with the following parameters,

python test.py -input_path ./data/hr_val/ \
               -output_path ./output/ \
               -encoder_name ResCNN \
               -pre_trained_model ./pre_trained_models/ArSSR_ResCNN.pkl \
               -scale 2 \
               -is_gpu 1 \
               -gpu 1

upon which the terminal outputs,

usage: test.py
       [-h]
       [-encoder ENCODER_NAME]
       [-depth DECODER_DEPTH]
       [-width DECODER_WIDTH]
       [-feature_dim FEATURE_DIM]
       [-pre_trained_model PRE_TRAINED_MODEL]
       [-is_gpu IS_GPU]
       [-gpu GPU]
       [-input_path INPUT_PATH]
       [-output_path OUTPUT_PATH]
       [-scale SCALE]
test.py: error: unrecognized arguments: -encoder_name ResCNN

The correct command has -encoder instead of -encoder_name. Please look into this issue.

自动分割的问题

你好，这是我得到的 GT：996782的分割图。感觉跟你论文里的差距还挺大。
我用这段代码获取的 segmentation.nii.gz， segmentation_dict = antspynet.utilities.deep_atropos(image, do_preprocessing=True)
分割网络推理时需要的模板图像用的是它自己下载的， brainExtractionRobustT1.h5，croppedMni152.nii.gz，croppedMni152Priors.nii.gz ， S_template3.nii.gz， sixTissueOctantBrainSegmentationWithPriors1.h5， 这里这些模板图像需要更改吗？

关于测试集下采样的问题

你好
我自己用的脚本，在{2，2.5，..} 进行下采样，用3d slicer观察， 数据维度是正常的，但图像空间不正确（0.7mm x 0.7mm x0.7mm）没有按照预计的变大。

你好，如何使用ANTs得到下面的这样的分割图？

HCP_1200 T1w原始数据

作者您好，我用您发的脚本下载的HCP_1200 t1w的图像数据，尺寸大小为（260，311，260），并不是您论文里尺寸为（320320256）您可以再提供一下HCP-1200 T1W的原始数据的网盘链接吗（之前的过期了）。第二个问题是，您是如何讲图片裁剪到264264264的？万分感谢您可以抽空回答我。邮箱：[email protected]

quality metrics

Hi，
Thanks for the code and paper. It is interesting to train a model for Arbitrary SR.
First one, I don't konw why the T1WIs that I download from HCP website, such as, 531536_3T_T1w_MPR1.nii.gz, are different from the 531536.nii.gz you provide. So，do you make some preprocessing for the original T1Ws MR images, such as, N4 bias correction and skull-stripping?

Second one, when I try to reproduce this experiment, there are some questions about quality metrics, LPIPS, PSI,LPC-SI. How to use the slice-by-slice strategy to compute them? Would mind sharing the code about this partment?

Look forward to any replies from you!
Best regards.

Question about evaluation metric calculation

Hi, thanks for uploading your codes. I enjoyed reading your paper and codes.
But I have one question.
After I have a predicted 3D images, how can I evaluate those with ground-truth images?
I see you have PSNR, SSIM, LPIPS, PSI, and LCP-SI.
Do you have a codes for this evaluation calculation? If so, can you also upload it by any chance?

Thanks!

training loss

尊敬的作者您好，你们做出了非常棒的工作，我在复现你们实验结果时，遇到了一个问题，训练 loss 几乎不会变化(一直维持在 0.3)，请问这个现象是正常的吗？希望得到您的解答！

About inference time

Hi
I have a question regarding the inference time for a sample with an HR dimension of 264264264 and a scale of 2. Could you please provide an estimate of how long the inference time would be?