sangyun884 / blur-diffusion Goto Github PK

Official PyTorch implementation of the paper Progressive Deblurring of Diffusion Models for Coarse-to-Fine Image Synthesis.

License: MIT License

Python 99.70% Shell 0.30%

diffusion-models generative-model image-generation image-synthesis score-based-generative-modeling score-based-generative-models score-based-models

blur-diffusion's People

Contributors

Stargazers

Watchers

Forkers

cv-synthesis multipath dl-diffusion zhuxiongwei24 georgepar cenkbircanoglu jackzhousz andyherko

blur-diffusion's Issues

AttributeError: 'EMA' object has no attribute 'defaults'`

Hello , thank you for the code.
I get a AttributeError when i want to train the model:
Traceback (most recent call last): File "main.py", line 292, in <module> optimizer.zero_grad() File "/home/extradisk/miniconda3/envs/sr3/lib/python3.8/site-packages/torch/optim/optimizer.py", line 231, in zero_grad foreach = self.defaults.get('foreach', False) AttributeError: 'EMA' object has no attribute 'defaults'
Could you help me with this bug ?

wonder why set N+1 interleaves

blur-diffusion/blur_diffusion.py

Line 40 in 9dd8adb

idxs = np.arange(N+1)

this will cause self.betas has N+1 length, which not consistent with other noise schedule settings

Have you normalize the input image to [-1,1]

I trained the model with medical images normalized to [0,1], and have achieved satisfactory results. I'd like to ask if it is necessary to normlize the image to [-1,1]?

LSUN bedroom test = LSUN bedroom train

Dear authors,

Is it intentional that the train and test datasets for the LSUN bedroom coincide? For other datasets in your code, this is not the case
If I understand correctly, this does not affect the FID calculations, but it just seems a bit odd.

blur-diffusion/main.py

Lines 100 to 110 in d86b07b

 elif dataset == 'lsun-bedroom': 

 res = 64 

 train_dir = "/home/ubuntu/code/sangyoon/forward-blur/datasets/bedroom_train/bedroom_train_lmdb/imgs" 

 dataset_train = dsets.LSUN(root='../forward-blur/datasets/bedroom_train', 

 classes=['bedroom_train'], 

 transform=transforms.Compose([transforms.Resize((res,res)), transforms.ToTensor()]) 

 ) 

 dataset_test = dsets.LSUN(root='../forward-blur/datasets/bedroom_train', 

 classes=['bedroom_train'], 

 transform=transforms.Compose([transforms.Resize((res,res)), transforms.ToTensor()]) 

 )

sangyun884 / blur-diffusion Goto Github PK

blur-diffusion's People

Contributors

Stargazers

Watchers

Forkers

blur-diffusion's Issues

AttributeError: 'EMA' object has no attribute 'defaults'`

wonder why set N+1 interleaves

Have you normalize the input image to [-1,1]

LSUN bedroom test = LSUN bedroom train

Recommend Projects

React

Vue.js

Typescript

TensorFlow

Django

Laravel

D3

Recommend Topics

javascript

web

server

Machine learning

Visualization

Game

Recommend Org

Facebook

Microsoft

Google

Alibaba

D3

Tencent

	elif dataset == 'lsun-bedroom':
	res = 64
	train_dir = "/home/ubuntu/code/sangyoon/forward-blur/datasets/bedroom_train/bedroom_train_lmdb/imgs"
	dataset_train = dsets.LSUN(root='../forward-blur/datasets/bedroom_train',
	classes=['bedroom_train'],
	transform=transforms.Compose([transforms.Resize((res,res)), transforms.ToTensor()])
	)
	dataset_test = dsets.LSUN(root='../forward-blur/datasets/bedroom_train',
	classes=['bedroom_train'],
	transform=transforms.Compose([transforms.Resize((res,res)), transforms.ToTensor()])
	)