MedSegDiff

MedSegDiff copied to clipboard

Described in https://github.com/MedicineToken/MedSegDiff/issues/193

pedropesserl

The generated image is black

2

The image I generated after training looks like this, is that correct please? How can I process it to get the image in the author's paper? Thank you so much!...

haoli-maker

How do I stop running code

2

I ran your code and found that lr_anneal_steps remained unchanged while the loop could not be stopped. Can you help me with this problem? Thank you 

helianggit

Hi, If I want to resume my interrupted training process, which checkpoint should I load?

kaneyxx

where is the ssformer?

3

i can not see the module of ss-former everywhere

qiiZzi

When I tried to run the mode on a cluster, I got this error (Binary operator not allowed in type expression [Ln 1764, Col 50]) & [Ln1831, Col 50]

Dr-LazyMazy

In version v1（MedSegDiff）, the dataset description states: *"We evaluated the performance of our method for glaucoma, thyroid cancer, and melanoma diagnosis on the REFUGE-2 dataset (Fang et al., 2022), BraTs-2021...

FionaYeager

How is the segmentation mask given to the model for multi label segmentation mask

4

If the segmentation mask has just two class background and cancer we can give it as mask with 1 and 0s If there are multiple classes, how are they given,...

nobleaustine

Liu-XL-code

一些问题的解答/Answers to some questions for training and sampling

13

看到这里很多issue没有人回答，作者也没怎么维护，我结合我自己实验的结果回答一下 1. 为什么采样生成的图案是一片漆黑？ 训练出了问题，大概率training过程的步数不够，多训练一会 2. training的epoch在多少比较合适 我个人实验的结果显示epoch大概在60000次左右效果开始变好 3. 代码没有设置收敛标准，怎么让training停下来 需要手动停止，停止的标准可以参考上一步我实验得出的结果 4. 代码生成了三个模型，用哪一个采用会比较好 emasavemodel.pt做采用的效果会比较好

MilkTeaAddicted