关于dpm_solver的使用问题

[xupinggl]

1、我用的作者给的ISIC数据集,其他的默认设置全部和作者一样，跑了10w step，但是sample的output基本是全黑的，或者是有一些随机噪声。有没有可能是我在train的时候将“--diffusion_steps 1000”改为了“--diffusion_steps 50 --dpm_solver True”，dpm_solver是只能在sampling的时候用吗？在train的时候用是不是不能用？

2、我在sampling的时候，也将“--diffusion_steps 1000”改为了“--diffusion_steps 50 --dpm_solver True”，推理时间是比较快，但是推理完两张后，就显示 cuda of memory

[heikeyuhuajia]

Ans1: dpm_solver 只在 sampling 上用。 对于 ISIC 数据集的问题等作者大大回复吧，我没用这个数据集； Ans2: dim-solver 只能在 PyTorch 1.8 上使用，作者还在解决。

[xupinggl]

@heikeyuhuajia thank you very much！

[DWBSIC]

Ans1: dpm_solver 只在 sampling 上用。 对于 ISIC 数据集的问题等作者大大回复吧，我没用这个数据集； Ans2: dim-solver 只能在 PyTorch 1.8 上使用，作者还在解决。

你好，我在Linux系统上使用dpm_solver是没有问题的，pytorch版本是1.9

[xupinggl]

@DWBSIC 好的，谢谢，我用的windows

[DWBSIC]

@DWBSIC 好的，谢谢，我用的windows

@xupinggl Traceback (most recent call last): File "D:\scripts\segmentation_sample.py", line 166, in main() File "D:\scripts\segmentation_sample.py", line 100, in main start = th.cuda.Event(enable_timing=True) File "D:\Program Files\Anaconda\envs\py39\lib\site-packages\torch\cuda\streams.py", line 164, in new return super(Event, cls).new( File "D:\Program Files\Anaconda\envs\py39\lib\site-packages\torch\cuda_utils.py", line 46, in err_fn raise RuntimeError( RuntimeError: Tried to instantiate dummy base class Event

您好，我在Linux系统中进行sample没有任何问题，但是我尝试在Windows系统下进行训练，所有的参数配置都是按照作者在readme中的参数设置，请问为什么会出现上述的错误。

Hello, I have no problem with sampling in Linux system, but I try to train in Windows system, and all parameters are set according to the author's parameter Settings in readme. May I ask why the above errors occur, please help to solve.

if you want to chat with me: WeChat: DWBSIC

[jaceqin]

Ans1: dpm_solver 只在 sampling 上用。 对于 ISIC 数据集的问题等作者大大回复吧，我没用这个数据集； Ans2: dim-solver 只能在 PyTorch 1.8 上使用，作者还在解决。

在pytorch 1.8上也有一样的BUG

[Issues-translate-bot]

Bot detected the issue body's language is not English, translate it automatically. 👯👭🏻🧑‍🤝‍🧑👫🧑🏿‍🤝‍🧑🏻👩🏾‍🤝‍👨🏿👬🏿

---

Ans1: dpm_solver is only used for sampling. Please wait for the author to reply to the question about the ISIC dataset. I don’t use this dataset; Ans2: dim-solver can only be used on PyTorch 1.8, and the author is still working on it.

There is the same BUG on pytorch 1.8

[xupinggl]

Has the problem been solved? I switched to linux, pytorch=1.8.0, and still got the error “cuda of memory”

[DWBSIC]

Has the problem been solved? I switched to linux, pytorch=1.8.0, and still got the error “cuda of memory”

是不是因为显卡内存不够用，或者是batch_size太大了

Is it because the graphics card memory is not enough, or the batch_size is too large

[WuJunde]

1. DPM is used for sampling and would cause performance compromise
2. same issue here: https://github.com/WuJunde/MedSegDiff/issues/69

[ipwjincheng]

1、我我用的作者给的ISIC数据集,其他的默认设置全部和作者一样，跑了10w step，但是sample的output基本是全黑的，或者是随机的噪声。是否是我在train的时候会“--diffusion_steps 1000”改为“--diffusion_steps 50--dpm_solver True”，dpm_solver是只能在sampling时用吗？在train的时候用是不是不能用？>> 2、我在sampling时，也将“--diffusion_steps 1000”改为“--diffusion_steps 50--dpm_solver True”推理时间是比较快，但推理完两张后，就显示 cuda of memory hello, when I use dim_solver, the inference image become very bad, have you met this problem

[xupinggl]

    您好，您的邮件我已收到，我会尽快回复的，谢谢！

[Issues-translate-bot]

Bot detected the issue body's language is not English, translate it automatically. 👯👭🏻🧑‍🤝‍🧑👫🧑🏿‍🤝‍🧑🏻👩🏾‍🤝‍👨🏿👬🏿

---

​ ​ Hello, I have received your email and I will reply as soon as possible, thank you!