一. AdaAttN-Revisit Attention Mechanism in Arbitrary Neural Style Transfer(ICCV2021)
- Download vgg_normalised.pth
- openvisdom
python -m visdom.server
- Configure content_path, style_path and image_encoder_path in train_adaattn.sh, which respectively represent the paths of the training content image, training style image and "vgg_normalised.pth" folder.
python train.py --content_path F:\RefDayDataset\KAIST_256\trainA --style_path F:\RefDayDataset\KAIST_256\trainB --name AdaAttN_kaist --model adaattn --dataset_mode unaligned --no_dropout --load_size 286 --crop_size 256 --image_encoder_path C:\Users\64883\Desktop\AdaAttN-main\models\vgg_normalised.pth --gpu_ids 0 --batch_size 1 --n_epochs 2 --n_epochs_decay 3 --display_freq 1 --display_port 8097 --display_env AdaAttN --lambda_local 3 --lambda_global 10 --lambda_content 0 --shallow_layer --skip_connection_3
Question 1
OSError: [WinError 1455] 页面文件太小,无法完成操作。 Error loading "D:\Anaconda3\envs\paddlepaddle\lib\site-packages\torch\lib\cudnn_cnn_infer64_8.dll" or one of its dependencies.
self._popen = self._Popen(self)
File "D:\Anaconda3\envs\paddlepaddle\lib\multiprocessing\context.py", line 223, in _Popen
return _default_context.get_context().Process._Popen(process_obj)
File "D:\Anaconda3\envs\paddlepaddle\lib\multiprocessing\context.py", line 322, in _Popen
return Popen(process_obj)
File "D:\Anaconda3\envs\paddlepaddle\lib\multiprocessing\popen_spawn_win32.py", line 89, in __init__
reduction.dump(process_obj, to_child)
File "D:\Anaconda3\envs\paddlepaddle\lib\multiprocessing\reduction.py", line 60, in dump
ForkingPickler(file, protocol).dump(obj)
BrokenPipeError: [Errno 32] Broken pipe
Solution
parser.add_argument('--num_threads', default=4, type=int, help='# threads for loading data')
change into
parser.add_argument('--num_threads', default=0, type=int, help='# threads for loading data')
Question 2
torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 4.08 GiB (GPU 0; 8.00 GiB total capacity; 134.76 MiB already allocated; 4.94 GiB free; 748.00 MiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for M
emory Management and PYTORCH_CUDA_ALLOC_CONF
Solution: Lower the resolution
Problem 3 The output frequency is too frequent
Solution
--display_freq 1
change to
--display_freq 1000
Problem 4 Content loss is always 0
Solution
--lambda_content 0
change into
--lambda_content 10
Problem 5: Too few training rounds
Solution
--n_epochs 2 --n_epochs_decay 3
change into
--n_epochs 100 --n_epochs_decay 100
二. ArtFlow- Unbiased Image Style Transfer via Reversible Neural Flows(CVPR2021)
- Download the VGG model, create a models folder, and move the model to the models folder
- Modify the training code
Create the experiments folder
python -u train.py --content_dir F:/RefDayDataset/KAIST_256/trainA --style_dir F:/RefDayDataset/KAIST_256/trainB --save_dir ./experiments/ArtFlow-AdaIN --n_flow 8 --n_block 2 --batch_size 4 --operator adain
Question 1
Traceback (most recent call last):
File "train.py", line 152, in <module>
content_dataset = FlatFolderDataset(args.content_dir, content_tf)
File "train.py", line 37, in __init__
self.paths = os.listdir(self.root)
OSError: [WinError 123] 文件名、目录名或卷标语法不正确。: "'F:\\RefDayDataset\\KAIST_256\\trainA'"
Solution: Remove the single quotes
Question 2
RuntimeError:
An attempt has been made to start a new process before the
current process has finished its bootstrapping phase.
This probably means that you are not using fork to start your
child processes and you have forgotten to use the proper idiom
in the main module:
if __name__ == '__main__':
freeze_support()
...
The "freeze_support()" line can be omitted if the program
is not going to be frozen to produce an executable.
Solution
parser.add_argument('--n_threads', type=int, default=8)
change into
parser.add_argument('--n_threads', type=int, default=0)
Question 3
torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 36.00 MiB (GPU 0; 8.00 GiB total capacity; 7.42 GiB already allocated; 0 bytes free; 7.47 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memor
y Management and PYTORCH_CUDA_ALLOC_CONF
Solution: Reduce batchsize and reduce resolution
--batch_size 4
change into
--batch_size 1
三. IEST- Artistic Style Transfer with Internal-external Learning and Contrastive Learning(NeurIPS2021)
- Download the VGG model and move it to the models folder
- Modify training code
python train.py --content_dir F:/RefDayDataset/KAIST_256/trainA --style_dir F:/RefDayDataset/KAIST_256/trainB
Question 1
RuntimeError:
An attempt has been made to start a new process before the
current process has finished its bootstrapping phase.
This probably means that you are not using fork to start your
child processes and you have forgotten to use the proper idiom
in the main module:
if __name__ == '__main__':
freeze_support()
...
The "freeze_support()" line can be omitted if the program
is not going to be frozen to produce an executable.
Solution
parser.add_argument('--n_threads', type=int, default=16)
change into
parser.add_argument('--n_threads', type=int, default=0)
Question 2
torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 36.00 MiB (GPU 0; 8.00 GiB total capacity; 7.42 GiB already allocated; 0 bytes free; 7.47 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memor
y Management and PYTORCH_CUDA_ALLOC_CONF
Solution: Reduce batchsize and reduce resolution
parser.add_argument('--batch_size', type=int, default=12)
change into
parser.add_argument('--batch_size', type=int, default=2)
Question 3
RuntimeError: CUDA error: device-side assert triggered
CUDA kernel errors might be asynchronously reported at some other API call,so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1.
Solution: Try on another card, or change num_workers
四. CAST- Domain Enhanced Arbitrary Image Style Transfer via Contrastive Learning(SIGGRAPH2022)
- Download pretrained style classification model and pretrained content encoder
- Modify training code
python train.py --dataroot F:/RefDayDataset/KAIST_256 --name cast
Question 1
File "<frozen importlib._bootstrap>", line 1006, in _gcd_import
File "<frozen importlib._bootstrap>", line 983, in _find_and_load
File "<frozen importlib._bootstrap>", line 967, in _find_and_load_unlocked
File "<frozen importlib._bootstrap>", line 677, in _load_unlocked
File "<frozen importlib._bootstrap_external>", line 728, in exec_module
File "<frozen importlib._bootstrap>", line 219, in _call_with_frames_removed
File "C:\Users\64883\Desktop\CAST_pytorch-main\models\cast_model.py", line 11, in <module>
import kornia.augmentation as K
ModuleNotFoundError: No module named 'kornia'
Solution
pip install kornia
Question 2
requests.exceptions.ConnectionError: HTTPConnectionPool(host='localhost', port=8097): Max retries exceeded with url: /env/main (Caused by NewConnectionError('<urllib3.connection.HTTPConnection object at 0x00000230810E0588>: Failed to establish a new connection: [WinError 10061] 由于目标计算机积极拒绝,无法连接。')
)
[WinError 10061] 由于目标计算机积极拒绝,无法连接。
on_close() takes 1 positional argument but 3 were given
Visdom python client failed to establish socket to get messages from the server. This feature is optional and can be disabled by initializing Visdom with `use_incoming_socket=False`, which will prevent waiting for this request to timeout.
Traceback (most recent call last):
File "D:\Anaconda3\envs\paddlepaddle\lib\site-packages\urllib3\util\connection.py", line 85, in create_connection
sock.connect(sa)
ConnectionRefusedError: [WinError 10061] 由于目标计算机积极拒绝,无法连接。
During handling of the above exception, another exception occurred:
Solution
python -m visdom.server
Question 3
Traceback (most recent call last):
File "<string>", line 1, in <module>
File "D:\Anaconda3\envs\paddlepaddle\lib\multiprocessing\spawn.py", line 105, in spawn_main
exitcode = _main(fd)
File "D:\Anaconda3\envs\paddlepaddle\lib\multiprocessing\spawn.py", line 115, in _main
self = reduction.pickle.load(from_parent)
EOFError: Ran out of input
Solution
parser.add_argument('--num_threads', default=4, type=int, help='# threads for loading data')
change into
parser.add_argument('--num_threads', default=0, type=int, help='# threads for loading data')
Question 4
torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 4.08 GiB (GPU 0; 8.00 GiB total capacity; 751.44 MiB already allocated; 4.37 GiB free; 1.30 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Mem
ory Management and PYTORCH_CUDA_ALLOC_CONF
Solution: Reduce batchsize and reduce resolution
五. StyTr2- Image Style Transfer with Transformers(CVPR2022)
- Download the VGG model and move it to the models folder
- Modify training code
python train.py --content_dir F:/RefDayDataset/KAIST_256/trainA --style_dir F:/RefDayDataset/KAIST_256/trainB --save_dir experiments/ --batch_size 1
Question 1
ImportError: cannot import name '_new_empty_tensor' from 'torchvision.ops' (D:\python\lib\site-packages\torchvision\ops\__init__.py)
Solution
import torchvision
if float(torchvision.__version__[:3]) < 0.7:
from torchvision.ops import _new_empty_tensor
from torchvision.ops.misc import _output_size
change into
import torchvision
if float(torchvision.__version__[2:4]) < 7:
from torchvision.ops import _new_empty_tensor
from torchvision.ops.misc import _output_size
Question 2
ImportError: cannot import name 'container_abcs' from 'torch._six' (D:\Anaconda3\envs\paddlepaddle\lib\site-packages\torch\_six.py)
Solution
from torch._six import container_abcs
change into
import collections.abc as container_abcs
Question 3
File "D:\Anaconda3\envs\paddlepaddle\lib\site-packages\torch\_utils.py", line 577, in <lambda>
return [_get_device_attr(lambda m: m.get_device_properties(i)) for i in device_ids]
File "D:\Anaconda3\envs\paddlepaddle\lib\site-packages\torch\cuda\__init__.py", line 374, in get_device_properties
raise AssertionError("Invalid device id")
AssertionError: Invalid device id
Solution
Comment out line 116 in train
# network = nn.DataParallel(network, device_ids=[0,1])
Question 4
RuntimeError:
An attempt has been made to start a new process before the
current process has finished its bootstrapping phase.
This probably means that you are not using fork to start your
child processes and you have forgotten to use the proper idiom
in the main module:
if __name__ == '__main__':
freeze_support()
...
The "freeze_support()" line can be omitted if the program
is not going to be frozen to produce an executable.
Solution
parser.add_argument('--n_threads', type=int, default=16)
change into
parser.add_argument('--n_threads', type=int, default=0)
Question 5
Traceback (most recent call last):
File "train.py", line 135, in <module>
{
'params': network.module.transformer.parameters()},
File "D:\Anaconda3\envs\paddlepaddle\lib\site-packages\torch\nn\modules\module.py", line 1270, in __getattr__
type(self).__name__, name))
AttributeError: 'StyTrans' object has no attribute 'module'
This error usually occurs when using PyTorch for multi-GPU training. In multi-GPU training, models are often wrapped in nn.DataParallel or nn.parallel.DistributedDataParallel to enable parallel computing. This results in changes to the model object's property access.
Solution
optimizer = torch.optim.Adam([
{
'params': network.module.transformer.parameters()},
{
'params': network.module.decode.parameters()},
{
'params': network.module.embedding.parameters()},
], lr=args.lr)
change to
optimizer = torch.optim.Adam([
{
'params': network.transformer.parameters()},
{
'params': network.decode.parameters()},
{
'params': network.embedding.parameters()},
], lr=args.lr)
六. QuantArt- Quantizing Image Style Transfer Towards High Visual Fidelity(CVPR2023)
- Create kaist.yaml
- Run the training code
python -u main.py --base configs/kaist.yaml -t True --gpus 0