chairc / integrated-design-diffusion-model Goto Github PK

Version: 1.1.3
Type: BUG

When I use unconditional to interrupt training, loading the model again will cause the following error.

Traceback (most recent call last): File "D:\Integrated-Design-Diffusion-Model\tools\train.py", line 408, in <module> main(args) File "D:\Integrated-Design-Diffusion-Model\tools\train.py", line 293, in main train(args=args) File "D:\Integrated-Design-Diffusion-Model\tools\train.py", line 156, in train load_ckpt(ckpt_path=pretrain_path, model=model, device=device, is_pretrain=pretrain, load_model_ckpt(model=model, model_ckpt=ckpt_model, is_train=is_train, is_pretrain=is_pretrain, File "D:\Integrated-Design-Diffusion-Model\utils\checkpoint.py", line 115, in load_model_ckpt model_weights_dict = {k: v for k, v in model_weights_dict.items() if np.shape(model_dict[k]) == np.shape(v)} File "D:\Integrated-Design-Diffusion-Model\utils\checkpoint.py", line 115, in <dictcomp> model_weights_dict = {k: v for k, v in model_weights_dict.items() if np.shape(model_dict[k]) == np.shape(v)} KeyError: 'label_emb.weight'

The model incorrectly loaded the weights in the conditional mode. I think there is an error here. Should we add conditional parameter to the load_model_ckpt() method in checkpoint.py?

This Issue is to summarize all kinds of problems and provide corresponding solutions. If there is no relevant problem in this issue, you can propose a new issue, and I will answer it.

Thanks for feedback bugs and contributing code (and pr).

这个Issue是对各种问题进行总结，并提供相应的解决方案。如果这个issue没有回答相关的问题，你可以提出一个新的issue，我会解答。

另外，感谢反馈bug和贡献代码(和pr)。

Problem quick navigation
问题快速导航
Q1. What is the purpose of this project? What significance does it have? (这个项目是做什么的？它有什么意义？)
Q2. How should I choose appropriate parameters during training? (我该如何在训练时选择合适的参数？)
Q3. How can I accelerate image generation during training? (我该如何在训练时加速图像生成？)
Q4. Why am I encountering numerous CUDA or cuDNN errors such as THCudaCheck FAIL file=../aten/src/THC/THCCachingHostAllocator.cpp or RuntimeError: cuDNN error: CUDNN_STATUS_INTERNAL_ERROR during training? (为什么我在训练的时候出现了THCudaCheck FAIL file=../aten/src/THC/THCCachingHostAllocator.cpp或RuntimeError: cuDNN error: CUDNN_STATUS_INTERNAL_ERROR等大片CUDA或cuDNN错误？)
Q5. Why do I see noise issues in the generated images? (为什么我生成的图片会出现噪点问题？)
Q6. How should the dataset be divided? How to set up conditional and unconditional training? (数据集该如何划分？条件训练和非条件训练该怎么设置？)
Q7. The training was interrupted unexpectedly. How can I resume training? (训练异常中断了，如何恢复训练？)
Q8. The training time for each epoch is too long. How can I use a pretrained model? (每轮训练时间太长了，怎么使用预训练模型？)
Q9. Why does using a 32×32 model to generate 64×64 or 128×128 images result in distortion and more objects? (为什么使用32×32的模型生成64×64的图片会扭曲、物体会变多呢？)
Q10. Why do I get a RuntimeError: Address already in use error when starting training? (为什么我启动训练报RuntimeError: Address already in use错误？)

你好，

想问一下你在训练cifar-10的时候的超参是什么样的呢？

大佬 我是用此模型来生成图像，输入是灰度图，在训练后 也能生成灰度图像吗？ 我跑了200epoch 后生成的是彩色图像

有什么好的路线吗读这个源码_(:з)∠)_, 新手看地比较杂乱.

我在完成epoch76的时候断网了, 恢复设置为start_epoch为77, 但是开始训练的时候显示是epoch 80... 虽然不影响什么性能但也提一下吧(

作者你好，小白想问一下，生成和训练的数据集是否要想训练其他神经网络一样划分数据集呀？

用的ddpm，训练的参数和生成的参数都是一样的，生成出来的是噪点图，想问下是哪部分出现了问题呢

在训练文件下，
parser.add_argument("--train_dataset_path", type=str,
default="G:/diffusion/Integrated-Design-Diffusion-Model-main/datasets/dataset_demo/class_1")
parser.add_argument("--val_dataset_path", type=str,
default="G:/diffusion/Integrated-Design-Diffusion-Model-main/datasets/dataset_demo/class_2")
# Enable automatic mixed precision training (needed)
我在训练集和验证集放入不同的图片，但是训练出来的图片总是和验证集一模一样，请问这是什么原因？

训练集不起作用，而且生成的图片，不同的轮次的高清hr图片跟原本验证集图片一样，但是可以看到训练过程中还有损失比如MSE=0.00471 求解答

If I do not reply in time, you can click this issue to find the solution.
如果没有及时回复，可先在此issue中查询

您好 首先感謝作者傑出的作品
我注意到你們conditional的條件是用類別來區分，不同於一般diffusion模型用text
且你們的模型輸入時也沒有prompt輸入

想請問你們的做法跟一般diffusion的差異 謝謝

作者大佬你好,
作者说40GiB显存足够训练128尺寸的图片.
请问作者团队有试过你们训练的最大可能吗? 比如256x256, 请问作者团队最大训练到多少尺寸, 用的配件条件大概是什么样的
非常感谢大佬！

When I trained 120 size images to epoch 30, nan appeared and all images turned black. I didn't save the model for each training, so I had to retrain my model.

I think we should add a way to prevent nan.

由于我的业务领域数据集比较难收集，所以我看了看ddpm，来生成图像数据集。但是看了看，感觉不太现实，生成质量和计算资源都比较难处理