sample得到预测的噪声后怎么去除噪声呢？

[qxxfd]

我的任务是实现分割，输入的图片对是原图 分割mask图，在使用readme的教程中训练的我的代码，结果发现 模型参数ModelMeanType.EPSILON: noise,也就是unet预测的noise么，调用segmentation_sample.py函数得到的
sample, x_noisy, org, cal, cal_out = sample_fn(
model,
(batch_size, channels, args.image_size, args.image_size), img,
step = args.diffusion_steps,
clip_denoised=args.clip_denoised,
model_kwargs=model_kwargs,
)
sample是noise么，之后怎么处理得到我要的target分割图呢
我在可视化sample后发现和我想要的mask相差甚远，看了模型参数发现预测的是noise 值，是否有朋友知道有输入图 noise 变量 怎么得到mask图，请教各位互联网朋友们，感谢！

[Issues-translate-bot]

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

---

Title: sampleHow to remove the noise after getting the predicted noise?

My task is to achieve segmentation. The input image pair is the original image segmentation mask map. I trained my code in the tutorial using the readme. It turns out that the model parameter ModelMeanType.EPSILON: noise, which is the noise predicted by unet, calls segmentation_sample. .py function obtained
sample, x_noisy, org, cal, cal_out = sample_fn(
model,
(batch_size, channels, args.image_size, args.image_size), img,
step = args.diffusion_steps,
clip_denoised=args.clip_denoised,
model_kwargs=model_kwargs,
)
Is the sample noise? How can I get the target segmentation map I want?
After visualizing the sample, I found that the mask was far from what I wanted. After looking at the model parameters, I found that the predicted noise value was. Does anyone know how to get the mask map with the noise variable in the input map? I would like to ask all Internet friends, thank you!

[hourlass]

Same question with you , have you solve your problem?

[invoker-qqwrv]

same question

[XiaowangjiAaa]

Same question with you , have you solve your problem?

[ashevimostar]

Same question with you , have you solve your problem?