zunzhumu / cpmnetv2 Goto Github PK

Hello, thank you for providing this code, and congratulations on your achievements. Could you please write down how to reproduce
your code results on LUNA16, such as the whole process of training and testing, data preprocessing and other codes (generated from LUNA16 original data)? Thank you very much.

请问--lambda_cls=4.0 --lambda_shape=0.1 --lambda_offset=1.0 --lambda_iou=1.0这几个参数在复现肺结节检测的时候怎么调节，论文里面好像没有写到，方便解释和提供一下吗？

I want to use it in kits2021 public data, how to build train and val csv?

你好，我好像没有找到luna16的训练文件，只找到了luna16_preprocessing.py，是否可能分享一份LUNA16的训练文件或者指导我如何进行训练呢？

Hello, I would like to reproduce your result on LUNA16, could you provide the data preparation script?

Thanks,

First of all, thank you for your willingness to share the code, it is very helpful to me. I am tyring to reproduce the result in luna16 and not hit the 0.956 froc.
So I am very interested in the hyperparameter settings on Luna16. Can you give me the corresponding hyperparameter settings?
Looking forward to your prompt reply! 😊
Ps: I know you're probably very busy, so I don't need the full Luna training code
@zunzhumu

Hi,

congratulations on your great work on CPMNetv2, this looks very interesting. The readme states that CPMNetv2 achieves very nice results on Aneurysm Detection on TOF-MRA (ADAM?) and RibFracture (RibFrac?) detection problems. Would it be possible to share those results and provide a citable source for them? (optimally also including the exact splits and experimental setup used to run those :) )

Best,
Michael

Hello, when I was training my own data, I found that the loss was still 0 when I reached the fifth epoch of training. Is there a problem with this? How to solve it?

您好，我在使用这个代码训练自己的数据时，出现这个错误：

RuntimeError: CUDA error: out of memory
CUDA kernel errors might be asynchronously reported at some other API call,so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1.

训练时我将batchsize和num_sam都改成1，显卡有一张A100, 9张RT3090，训练数据有712例，这个需要如何优化才可以跑起来呢？

您好，我在使用您的代码训练自己数据时，在train_CPMv2_RibFrac.sh文件添加root具体路径（--root = "/home/raid2/anne/dataset/nodels/"），当我运行train_CPMv2_RibFrac.sh文件时，为什么显示arge.root为空呢？

It's a good job that CPM-Net can get excellent perfermance on LUNA16. Recently, I wanted to use CPM-NetV2 to reproduce this experiment, but I got failed. The performance of my CPM-NetV2 model is not satisfying (its CPM is only 85.0). Could you please provide me with some suggestions on hyperparameters?