Ryann

@malonzia Thanks for your answer. So, does it mean that any type of data augmentation can be applied to the input image?

@xin-wo-1 I hope these two methods can help you solve the NaN problems: 1. set `model.precision=32`; 2. check the way of data preprocessing. And I'm also troubled by the white...

@xin-wo-1 Thanks for your answer. Since the reconstruction image isn't shown in the paper, I continued to train the DDPM model using my custom dataset due to time constraints. The...

@xin-wo-1 Hi there, the size of my input image is 256×256×32(h,w,d) and the latent vector encoded by the VQGAN I trained is of size 8×64×64×8(c,h,w,d). So I set `model.diffusion_img_size=64 model.diffusion_depth_size=8...

@xin-wo-1 I set `model.batch_size=10` using **two** NVIDIA RTX 3090 with 24GB GPU RAM, resulting in a total batch size of **20**. The loss gets stable at step 150k just as...

@xin-wo-1 I didn't change the timesteps in my experiments. I learned little from MRI. Could you give some examples of the missing MRI radiological features?

@kanydao Here's the official code: ```python """Adapted from https://github.com/peterhan91/cycleGAN/blob/db8f1d958c0879c29cf3932cae74a166317be812/prepro.py#L39""" import os import numpy as np from glob import glob import pydicom import scipy.ndimage from pathlib import Path import argparse from...

> 我有遇到这个问题了，但是我好像没有安装vllm，这个应该怎么回退呢？ @Kegard Hi, I ran into the same issue. Downgrading `aiohttp` from 3.12.x to **3.9.5** solved the problem.

Hi, I've encountered the same issue. Did you solve this yet?

> Hi, I've encountered the same issue. Did you solve this yet? I solve this issue by changing the usage of GPUs. ### Bash scripts that fail to run: *...