podgorskiy / alae Goto Github PK
View Code? Open in Web Editor NEW[CVPR2020] Adversarial Latent Autoencoders
[CVPR2020] Adversarial Latent Autoencoders
I added 9 images in the src and dst folder of /ALAE/style_mixing/test_images/set_ffhq/dst
updated src_len and dst_len in the code.
But got following error.
2020-04-29 22:11:35,659 logger INFO: Trainable parameters generator: 2020-04-29 22:11:35,660 logger INFO: Trainable parameters discriminator: 2020-04-29 22:11:35,660 logger INFO: Loading checkpoint from training_artifacts/ffhq/model_157.pth 2020-04-29 22:11:35,918 logger INFO: Model trained for 157 epochs Traceback (most recent call last): File "style_mixing/stylemix.py", line 192, in <module> world_size=gpu_count, write_log=False) File "/home/ubuntu/projects/ALAE/launcher.py", line 131, in run _run(0, world_size, fn, defaults, write_log, no_cuda, args) File "/home/ubuntu/projects/ALAE/launcher.py", line 96, in _run fn(**matching_args) File "style_mixing/stylemix.py", line 42, in main _main(cfg, logger) File "style_mixing/stylemix.py", line 180, in _main style = mix_styles(src_latents, row_latents, style_ranges[row]) IndexError: list index out of range
Line 129 in 5d8362f
I get this error when backpropagating as line below.
RuntimeError: cuDNN error: CUDNN_STATUS_NOT_SUPPORTED. This error may appear if you passed in a non-contiguous input.
Line 298 in 5d8362f
I installed all the requirements, and used PyTorch v1.4.0.
I included contiguous() everywhere after .view()
, .reshape()
, so I might not be non-contiguous problem.
I'm currently using single GPU, so it might not be multi-GPU problem.
Do you have any idea about this error?
You've reached a quota on your files, and dlutils doesn't throw an error.
Not really an issue with ALAE.
There's an easy fix : one can download the pretrained weights manually here : https://drive.google.com/drive/folders/1iZodDA4q1IKRRgV2nJuAyyuCwQGtL4vp?usp=sharing
How long did it take to train the style_alae 1024
Is it intentional that the D module (MappingToLatent) consists of three F.Linear layers w/o any activations (e.g. no ReLU / Leaky ReLU)?
Line 894 in 5d8362f
How much vram is needed to train the 1024 from scratch?
I'm sorry I know this is not a place for this but it is the only way I knew I how to contact.
In the Paper
It is written "possible with SyleGAN alone," 2 lines above Acknowledgments.
Nice work BTW.
Is it possible to support the functionality where we can run the interactive demo on a remote server (similar to tensorboard)? I have GPUs available only on a headless server, which might be the case for many others.
Thanks.
Just wanted to see if this is working or not
Hi,
Thanks for releasing such a well-written code and the interactive demo for the paper. Even when I am testing it on real-world images, the reconstruction and semantic changes are working very well.
However, I was wondering whether you are planning to release more details on the semantic editing part. In particular, I couldn't find the details on how the principal direction vectors are calculated in the paper. Surprisingly, the paper doesn't have any results on semantic editing, i.e., the ones demonstrated in the demo.
I wonder whether you are planning to release any additional documents including these details? Or is it a generic methodology well-known in the community? I am relatively new in this domain and not sure about it.
Thanks.
When running the interactive demo I get:
ERROR: ImGui_ImplOpenGL3_CreateDeviceObjects: failed to compile fragment shader!
ERROR: ImGui_ImplOpenGL3_CreateDeviceObjects: failed to link shader program!
Output of original StyleGAN's discriminator is a scalar, predicting whether the given image is real or fake. However, the output shape of your D network is batch x (2 * dlatent_size) in the line below.
Line 893 in 5d8362f
Line 111 in 5d8362f
I'm curious why the output shape of D network is batch x (2 * dlatent_size), since only one element is used for training and the others are useless.
Plus, I can't understand why the output of D network is reshaped like this.
Line 903 in 5d8362f
Windows 10, Python 3.6, all the dependencies installed
python interactive_demo.py
Is working for a few seconds and then exiting with no visual result or prompt. Any ideas on how to debug?
Line 283 in 5d8362f
Unable to generate female version of custom image.
Can any one help me?
Thank you in advance.
Reconstructed image of custom image is image of completely different person with same default settings.
I've used the command %pip install tensorflow-gpu==1.10
following your ReadMe.
The command is run after
%pip install -r requirements.txt
%pip install dareblopy
Then I copied the find_principal_directions.py to the Colab like this:
from dataloader import *
import numpy as np
import tensorflow as tf
import principal_directions.classifier
def parse_tfrecord_np(record):
ex = tf.train.Example()
ex.ParseFromString(record)
shape = ex.features.feature['shape'].int64_list.value
data = ex.features.feature['data'].bytes_list.value[0]
dlat = ex.features.feature['dlat'].bytes_list.value[0]
lat = ex.features.feature['lat'].bytes_list.value[0]
return np.fromstring(data, np.uint8).reshape(shape), np.fromstring(dlat, np.float32), np.fromstring(lat, np.float32)
class Predictions:
def __init__(self, cfg, minibatch_gpu):
self.minibatch_size = minibatch_gpu
self.cfg = cfg
def evaluate(self, logger, mapping, decoder, lod, attrib_idx):
result_expr = []
rnd = np.random.RandomState(5)
with tf.Graph().as_default(), tf.Session() as sess:
ds = tf.data.TFRecordDataset("principal_directions/generated_data.000")
ds = ds.batch(self.minibatch_size)
batch = ds.make_one_shot_iterator().get_next()
classifier = principal_directions.classifier.make_classifier(attrib_idx)
i = 0
while True:
try:
records = sess.run(batch)
images = []
dlats = []
lats = []
for r in records:
im, dlat, lat = parse_tfrecord_np(r)
# plt.imshow(im.transpose(1, 2, 0), interpolation='nearest')
# plt.show()
images.append(im)
dlats.append(dlat)
lats.append(lat)
images = np.stack(images)
dlats = np.stack(dlats)
lats = np.stack(lats)
logits = classifier.run(images, None, num_gpus=1, assume_frozen=True)
logits = torch.tensor(logits)
predictions = torch.softmax(torch.cat([logits, -logits], dim=1), dim=1)
result_dict = dict(latents=lats, dlatents=dlats)
result_dict[attrib_idx] = predictions.cpu().numpy()
result_expr.append(result_dict)
i += 1
except tf.errors.OutOfRangeError:
break
results = {key: np.concatenate([value[key] for value in result_expr], axis=0) for key in result_expr[0].keys()}
np.save("principal_directions/wspace_att_%d" % attrib_idx, results)
def main(cfg, logger):
torch.cuda.set_device(0)
model = Model(
startf=cfg.MODEL.START_CHANNEL_COUNT,
layer_count=cfg.MODEL.LAYER_COUNT,
maxf=cfg.MODEL.MAX_CHANNEL_COUNT,
latent_size=cfg.MODEL.LATENT_SPACE_SIZE,
truncation_psi=cfg.MODEL.TRUNCATIOM_PSI,
truncation_cutoff=cfg.MODEL.TRUNCATIOM_CUTOFF,
mapping_layers=cfg.MODEL.MAPPING_LAYERS,
channels=cfg.MODEL.CHANNELS,
generator=cfg.MODEL.GENERATOR,
encoder=cfg.MODEL.ENCODER)
model.cuda(0)
model.eval()
model.requires_grad_(False)
decoder = model.decoder
encoder = model.encoder
mapping_tl = model.mapping_tl
mapping_fl = model.mapping_fl
dlatent_avg = model.dlatent_avg
logger.info("Trainable parameters generator:")
count_parameters(decoder)
logger.info("Trainable parameters discriminator:")
count_parameters(encoder)
arguments = dict()
arguments["iteration"] = 0
model_dict = {
'discriminator_s': encoder,
'generator_s': decoder,
'mapping_tl_s': mapping_tl,
'mapping_fl_s': mapping_fl,
'dlatent_avg': dlatent_avg
}
checkpointer = Checkpointer(cfg,
model_dict,
{},
logger=logger,
save=False)
checkpointer.load()
model.eval()
layer_count = cfg.MODEL.LAYER_COUNT
logger.info("Extracting attributes")
decoder = nn.DataParallel(decoder)
indices = [0, 1, 2, 3, 4, 10, 11, 17, 19]
with torch.no_grad():
p = Predictions(cfg, minibatch_gpu=4)
for i in indices:
p.evaluate(logger, mapping_fl, decoder, cfg.DATASET.MAX_RESOLUTION_LEVEL - 2, i)
if __name__ == "__main__":
gpu_count = 1
run(main, get_cfg_defaults(), description='StyleGAN', default_config='configs/celeba.yaml',
world_size=gpu_count, write_log=False)
Here are the logs when I run each of the code blocks respectively:
Collecting tensorflow-gpu==1.10
Downloading https://files.pythonhosted.org/packages/64/ca/830b7cedb073ae264d215d51bd18d7cff7a2a47e39d79f6fa23edae17bb2/tensorflow_gpu-1.10.0-cp36-cp36m-manylinux1_x86_64.whl (253.2MB)
|โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ| 253.3MB 52kB/s
Requirement already satisfied: astor>=0.6.0 in /usr/local/lib/python3.6/dist-packages (from tensorflow-gpu==1.10) (0.8.1)
Collecting numpy<=1.14.5,>=1.13.3
Downloading https://files.pythonhosted.org/packages/68/1e/116ad560de97694e2d0c1843a7a0075cc9f49e922454d32f49a80eb6f1f2/numpy-1.14.5-cp36-cp36m-manylinux1_x86_64.whl (12.2MB)
|โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ| 12.2MB 38.8MB/s
Requirement already satisfied: termcolor>=1.1.0 in /usr/local/lib/python3.6/dist-packages (from tensorflow-gpu==1.10) (1.1.0)
Requirement already satisfied: wheel>=0.26 in /usr/local/lib/python3.6/dist-packages (from tensorflow-gpu==1.10) (0.34.2)
Requirement already satisfied: gast>=0.2.0 in /usr/local/lib/python3.6/dist-packages (from tensorflow-gpu==1.10) (0.3.3)
Requirement already satisfied: protobuf>=3.6.0 in /usr/local/lib/python3.6/dist-packages (from tensorflow-gpu==1.10) (3.12.4)
Requirement already satisfied: grpcio>=1.8.6 in /usr/local/lib/python3.6/dist-packages (from tensorflow-gpu==1.10) (1.31.0)
Collecting tensorboard<1.11.0,>=1.10.0
Downloading https://files.pythonhosted.org/packages/c6/17/ecd918a004f297955c30b4fffbea100b1606c225dbf0443264012773c3ff/tensorboard-1.10.0-py3-none-any.whl (3.3MB)
|โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ| 3.3MB 44.2MB/s
Collecting setuptools<=39.1.0
Downloading https://files.pythonhosted.org/packages/8c/10/79282747f9169f21c053c562a0baa21815a8c7879be97abd930dbcf862e8/setuptools-39.1.0-py2.py3-none-any.whl (566kB)
|โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ| 573kB 42.4MB/s
Requirement already satisfied: six>=1.10.0 in /usr/local/lib/python3.6/dist-packages (from tensorflow-gpu==1.10) (1.15.0)
Requirement already satisfied: absl-py>=0.1.6 in /usr/local/lib/python3.6/dist-packages (from tensorflow-gpu==1.10) (0.9.0)
Requirement already satisfied: markdown>=2.6.8 in /usr/local/lib/python3.6/dist-packages (from tensorboard<1.11.0,>=1.10.0->tensorflow-gpu==1.10) (3.2.2)
Requirement already satisfied: werkzeug>=0.11.10 in /usr/local/lib/python3.6/dist-packages (from tensorboard<1.11.0,>=1.10.0->tensorflow-gpu==1.10) (1.0.1)
Requirement already satisfied: importlib-metadata; python_version < "3.8" in /usr/local/lib/python3.6/dist-packages (from markdown>=2.6.8->tensorboard<1.11.0,>=1.10.0->tensorflow-gpu==1.10) (1.7.0)
Requirement already satisfied: zipp>=0.5 in /usr/local/lib/python3.6/dist-packages (from importlib-metadata; python_version < "3.8"->markdown>=2.6.8->tensorboard<1.11.0,>=1.10.0->tensorflow-gpu==1.10) (3.1.0)
ERROR: xarray 0.15.1 has requirement numpy>=1.15, but you'll have numpy 1.14.5 which is incompatible.
ERROR: xarray 0.15.1 has requirement setuptools>=41.2, but you'll have setuptools 39.1.0 which is incompatible.
ERROR: umap-learn 0.4.6 has requirement numpy>=1.17, but you'll have numpy 1.14.5 which is incompatible.
ERROR: tifffile 2020.7.24 has requirement numpy>=1.15.1, but you'll have numpy 1.14.5 which is incompatible.
ERROR: tensorflow 2.3.0 has requirement numpy<1.19.0,>=1.16.0, but you'll have numpy 1.14.5 which is incompatible.
ERROR: tensorflow 2.3.0 has requirement tensorboard<3,>=2.3.0, but you'll have tensorboard 1.10.0 which is incompatible.
ERROR: spacy 2.2.4 has requirement numpy>=1.15.0, but you'll have numpy 1.14.5 which is incompatible.
ERROR: plotnine 0.6.0 has requirement numpy>=1.16.0, but you'll have numpy 1.14.5 which is incompatible.
ERROR: numba 0.48.0 has requirement numpy>=1.15, but you'll have numpy 1.14.5 which is incompatible.
ERROR: imgaug 0.2.9 has requirement numpy>=1.15.0, but you'll have numpy 1.14.5 which is incompatible.
ERROR: google-auth 1.17.2 has requirement setuptools>=40.3.0, but you'll have setuptools 39.1.0 which is incompatible.
ERROR: fastai 1.0.61 has requirement numpy>=1.15, but you'll have numpy 1.14.5 which is incompatible.
ERROR: datascience 0.10.6 has requirement folium==0.2.1, but you'll have folium 0.8.3 which is incompatible.
ERROR: cvxpy 1.0.31 has requirement numpy>=1.15, but you'll have numpy 1.14.5 which is incompatible.
ERROR: blis 0.4.1 has requirement numpy>=1.15.0, but you'll have numpy 1.14.5 which is incompatible.
ERROR: astropy 4.0.1.post1 has requirement numpy>=1.16, but you'll have numpy 1.14.5 which is incompatible.
ERROR: albumentations 0.1.12 has requirement imgaug<0.2.7,>=0.2.5, but you'll have imgaug 0.2.9 which is incompatible.
Installing collected packages: numpy, tensorboard, setuptools, tensorflow-gpu
Found existing installation: numpy 1.18.5
Uninstalling numpy-1.18.5:
Successfully uninstalled numpy-1.18.5
Found existing installation: tensorboard 2.3.0
Uninstalling tensorboard-2.3.0:
Successfully uninstalled tensorboard-2.3.0
Found existing installation: setuptools 49.2.0
Uninstalling setuptools-49.2.0:
Successfully uninstalled setuptools-49.2.0
Successfully installed numpy-1.14.5 setuptools-39.1.0 tensorboard-1.10.0 tensorflow-gpu-1.10.0
WARNING: The following packages were previously imported in this runtime:
[numpy,pkg_resources]
You must restart the runtime in order to use newly installed versions.
[autoreload of pkg_resources._vendor.six failed: Traceback (most recent call last):
File "/usr/local/lib/python3.6/dist-packages/IPython/extensions/autoreload.py", line 247, in check
superreload(m, reload, self.old_objects)
AttributeError: 'NoneType' object has no attribute 'cStringIO'
]
/usr/local/lib/python3.6/dist-packages/numpy/add_newdocs.py:882: UserWarning: add_newdoc was used on a pure-python object <function empty_like at 0x7fd680b3e7b8>. Prefer to attach it directly to the source.
""")
/usr/local/lib/python3.6/dist-packages/numpy/add_newdocs.py:1239: UserWarning: add_newdoc was used on a pure-python object <function concatenate at 0x7fd680b3e8c8>. Prefer to attach it directly to the source.
""")
/usr/local/lib/python3.6/dist-packages/numpy/add_newdocs.py:1313: UserWarning: add_newdoc was used on a pure-python object <function inner at 0x7fd680b3e9d8>. Prefer to attach it directly to the source.
""")
/usr/local/lib/python3.6/dist-packages/numpy/add_newdocs.py:1519: UserWarning: add_newdoc was used on a pure-python object <function where at 0x7fd680b3eae8>. Prefer to attach it directly to the source.
""")
/usr/local/lib/python3.6/dist-packages/numpy/add_newdocs.py:1596: UserWarning: add_newdoc was used on a pure-python object <function lexsort at 0x7fd680b3ebf8>. Prefer to attach it directly to the source.
""")
/usr/local/lib/python3.6/dist-packages/numpy/add_newdocs.py:1704: UserWarning: add_newdoc was used on a pure-python object <function can_cast at 0x7fd680b3ed08>. Prefer to attach it directly to the source.
""")
/usr/local/lib/python3.6/dist-packages/numpy/add_newdocs.py:1804: UserWarning: add_newdoc was used on a pure-python object <function min_scalar_type at 0x7fd680b3ee18>. Prefer to attach it directly to the source.
""")
/usr/local/lib/python3.6/dist-packages/numpy/add_newdocs.py:1873: UserWarning: add_newdoc was used on a pure-python object <function result_type at 0x7fd680b3ef28>. Prefer to attach it directly to the source.
""")
[autoreload of numpy failed: Traceback (most recent call last):
File "/usr/local/lib/python3.6/dist-packages/IPython/extensions/autoreload.py", line 247, in check
superreload(m, reload, self.old_objects)
AttributeError: module 'numpy.core.multiarray' has no attribute 'newbuffer'
]
[autoreload of numpy.core failed: Traceback (most recent call last):
File "/usr/local/lib/python3.6/dist-packages/IPython/extensions/autoreload.py", line 247, in check
superreload(m, reload, self.old_objects)
ImportError: cannot import name '_numpy_tester'
]
[autoreload of numpy.core.numerictypes failed: Traceback (most recent call last):
File "/usr/local/lib/python3.6/dist-packages/IPython/extensions/autoreload.py", line 247, in check
superreload(m, reload, self.old_objects)
IndexError: string index out of range
]
[autoreload of numpy.core.numeric failed: Traceback (most recent call last):
File "/usr/local/lib/python3.6/dist-packages/IPython/extensions/autoreload.py", line 247, in check
superreload(m, reload, self.old_objects)
ImportError: cannot import name 'TooHardError'
]
[autoreload of numpy.lib failed: Traceback (most recent call last):
File "/usr/local/lib/python3.6/dist-packages/IPython/extensions/autoreload.py", line 247, in check
superreload(m, reload, self.old_objects)
NameError: name 'type_check' is not defined
]
[autoreload of numpy.matrixlib failed: Traceback (most recent call last):
File "/usr/local/lib/python3.6/dist-packages/IPython/extensions/autoreload.py", line 247, in check
superreload(m, reload, self.old_objects)
NameError: name 'defmatrix' is not defined
]
[autoreload of numpy.linalg failed: Traceback (most recent call last):
File "/usr/local/lib/python3.6/dist-packages/IPython/extensions/autoreload.py", line 247, in check
superreload(m, reload, self.old_objects)
ImportError: cannot import name '_numpy_tester'
]
[autoreload of numpy.lib.function_base failed: Traceback (most recent call last):
File "/usr/local/lib/python3.6/dist-packages/IPython/extensions/autoreload.py", line 247, in check
superreload(m, reload, self.old_objects)
ImportError: cannot import name 'digitize'
]
[autoreload of numpy.fft failed: Traceback (most recent call last):
File "/usr/local/lib/python3.6/dist-packages/IPython/extensions/autoreload.py", line 247, in check
superreload(m, reload, self.old_objects)
ImportError: cannot import name '_FFTCache'
]
[autoreload of numpy.polynomial failed: Traceback (most recent call last):
File "/usr/local/lib/python3.6/dist-packages/IPython/extensions/autoreload.py", line 247, in check
superreload(m, reload, self.old_objects)
ImportError: cannot import name '_numpy_tester'
]
[autoreload of numpy.random failed: Traceback (most recent call last):
File "/usr/local/lib/python3.6/dist-packages/IPython/extensions/autoreload.py", line 247, in check
superreload(m, reload, self.old_objects)
ImportError: cannot import name '_numpy_tester'
]
[autoreload of numpy.ma failed: Traceback (most recent call last):
File "/usr/local/lib/python3.6/dist-packages/IPython/extensions/autoreload.py", line 247, in check
superreload(m, reload, self.old_objects)
ImportError: cannot import name '_numpy_tester'
]
[autoreload of numpy.ma.core failed: Traceback (most recent call last):
File "/usr/local/lib/python3.6/dist-packages/IPython/extensions/autoreload.py", line 247, in check
superreload(m, reload, self.old_objects)
AttributeError: module 'numpy' has no attribute 'rank'
]
---------------------------------------------------------------------------
ImportError Traceback (most recent call last)
/usr/local/lib/python3.6/dist-packages/tensorflow/python/pywrap_tensorflow.py in <module>()
57
---> 58 from tensorflow.python.pywrap_tensorflow_internal import *
59 from tensorflow.python.pywrap_tensorflow_internal import __version__
7 frames
ImportError: libcublas.so.9.0: cannot open shared object file: No such file or directory
During handling of the above exception, another exception occurred:
ImportError Traceback (most recent call last)
/usr/local/lib/python3.6/dist-packages/tensorflow/python/pywrap_tensorflow.py in <module>()
72 for some common reasons and solutions. Include the entire stack trace
73 above this error message when asking for help.""" % traceback.format_exc()
---> 74 raise ImportError(msg)
75
76 # pylint: enable=wildcard-import,g-import-not-at-top,unused-import,line-too-long
ImportError: Traceback (most recent call last):
File "/usr/local/lib/python3.6/dist-packages/tensorflow/python/pywrap_tensorflow.py", line 58, in <module>
from tensorflow.python.pywrap_tensorflow_internal import *
File "/usr/local/lib/python3.6/dist-packages/tensorflow/python/pywrap_tensorflow_internal.py", line 28, in <module>
_pywrap_tensorflow_internal = swig_import_helper()
File "/usr/local/lib/python3.6/dist-packages/tensorflow/python/pywrap_tensorflow_internal.py", line 24, in swig_import_helper
_mod = imp.load_module('_pywrap_tensorflow_internal', fp, pathname, description)
File "/usr/lib/python3.6/imp.py", line 243, in load_module
return load_dynamic(name, filename, file)
File "/usr/lib/python3.6/imp.py", line 343, in load_dynamic
return _load(spec)
ImportError: libcublas.so.9.0: cannot open shared object file: No such file or directory
Failed to load the native TensorFlow runtime.
See https://www.tensorflow.org/install/install_sources#common_installation_problems
for some common reasons and solutions. Include the entire stack trace
above this error message when asking for help.
---------------------------------------------------------------------------
NOTE: If your import is failing due to a missing package, you can
manually install dependencies using either !pip or !apt.
To view examples of installing some common dependencies, click the
"Open Examples" button below.
---------------------------------------------------------------------------
Thank you for your great work! However, I'm having trouble with Bimpy.
I got segmentation fault error message on line 184 of interactive_demo.py.
Line 184 in 471301a
Segmentation fault error seems to appear by ctx.init as i tried simple example below. Do you have any idea about this case? I'm trying on Linux.
I'm trying to trying on MNIST and hitting the above error. Here's the backtrace:
Traceback (most recent call last):
File "train_alae.py", line 352, in <module>
run(train, get_cfg_defaults(), description='StyleGAN', default_config='configs/mnist.yaml',
File "/home/james/src/spliqsml/ALAE/launcher.py", line 131, in run
_run(0, world_size, fn, defaults, write_log, no_cuda, args)
File "/home/james/src/spliqsml/ALAE/launcher.py", line 96, in _run
fn(**matching_args)
File "train_alae.py", line 186, in train
scheduler = ComboMultiStepLR(optimizers=
File "/home/james/src/spliqsml/ALAE/scheduler.py", line 91, in __init__
self.schedulers[name] = WarmupMultiStepLR(opt, lr=base_lr, **kwargs)
File "/home/james/src/spliqsml/ALAE/scheduler.py", line 52, in __init__
self.step(last_epoch)
File "/home/james/anaconda3/envs/tf2/lib/python3.8/site-packages/torch/optim/lr_scheduler.py", line 166, in step
self.print_lr(self.verbose, i, lr, epoch)
AttributeError: 'WarmupMultiStepLR' object has no attribute 'verbose'
Kinda strange, since it seems like that's in your code, but called by torch's lr_scheduler... ??
tensorflow 2.3.1
libcudart.so.10.1
Went through all the setup on new conda env and ran with: python train_alae.py -c mnist
Any help appreciated.
Hey, the paper is amazing. I really loved it. I am trying to recreate the MNIST experiment that you did.
I only have Google CoLab as compute resource.
I prepared mnist TFRecords using available prepare_mnist_tfrecords.py
. The created data
dir is in the root directory.
When I try to train with mnist_fc.yaml
as config file, training seems to start. But after showing this log nothing happens.
I tried to debug the issue,
batches = make_dataloader(cfg, logger, dataset, lod2batch.get_per_GPU_batch_size(), local_rank)
After this line print('debug')
is not printing debug. Inside make_dataloader
function. The same is not happening after,
batches = db.data_loader(iter(dataset), BatchCollator(local_rank), len(dataset) // GPU_batch_size)
Am I not providing something to the training script? I have installed dareblopy
in my runtime.
To reproduce the experiment you can check out this notebook.
What's happening? Any suggestion.
Thank you.
I am trying to recreate the results from your permutation invariant MNIST section. Can you help me understand which all files should I use?
For example which file should I use between train_alae.py
and train_alae_separate.py
?
Similary between model.py
and model_separate.py
?
Thanks in advance @podgorskiy :)
train_alae.py seems to crash from time to time with the below error:
902it [04:31, 3.29it/s]Exception ignored in: <function Image.del at 0x7ff8eb6549d8>
Traceback (most recent call last):
File "/usr/lib/python3.7/tkinter/init.py", line 3507, in del
self.tk.call('image', 'delete', self.name)
RuntimeError: main thread is not in main loop
Tcl_AsyncDelete: async handler deleted by the wrong thread
Abort
As a workaround, I deleted the line @utils.async_func before def save_pic(x_rec) and this seems to solve the problem. It looks like some thread safety issue with the PIL Image destructor.
Update: I still got a crash with the workaround.
Hi,
Congratulations & thanks for this paper! I am running interactive_demo on colab and getting non-stop error
ERROR: ImGui_ImplOpenGL3_CreateDeviceObjects: failed to compile fragment shader! ERROR: ImGui_ImplOpenGL3_CreateDeviceObjects: failed to link shader program! ERROR: ImGui_ImplOpenGL3_CreateDeviceObjects: failed to compile vertex shader! ERROR: ImGui_ImplOpenGL3_CreateDeviceObjects: failed to compile fragment shader! ERROR: ImGui_ImplOpenGL3_CreateDeviceObjects: failed to link shader program! ERROR: ImGui_ImplOpenGL3_CreateDeviceObjects: failed to compile vertex shader! ERROR: ImGui_ImplOpenGL3_CreateDeviceObjects: failed to compile fragment shader!
Any solution?
Greetings!
I've been trying to train on MNIST (I have custom data that's MNIST-like) but keep hitting Segmentation fault (core dumped)
.
tensorflow-gpu = 1.15
pytorch = 1.4.0
I have doreblopy installed.
Out of curiosity I ran the interactive demo, which works fine.
0
Hello, I am about to find some new directions in W space. But, what is interactive_slider.py
in principal_directions/README.md
? I didn't find such a file in the repository. Does anybody know? Many thanks.
the error as in the title. Any thoughts on this?
Windows 10, Anaconda, PyTorch 1.4.0, CUDA 10.1 (for GeForce RTX 2080 with Max-Q Design), Visual Studio C++ 14.0 (for bimpy complication).
I run without any errors, yet the screen presents no results:
The narrow vertical line is not expandable.
EDIT:
On python interactive_demo.py -c celeba-hq256
I've found it is possible to find the lower right corner and resize it manually. On the original one, it is out of my reach.
A big advantage of StyleGAN is the seamless fine-tuning process, where a previous checkpoint can be used as a starting point for training on a new dataset (say for example fine-tune the FFHQ model on paintings).
Is this possible for ALAE too? Do you have any pointers or feedback on how to approach it?
Please add a license file.
Hi,
Could you please help with startup of training?
After the start of training script I get this output:
_2020-07-06 19:11:33,163 logger INFO: Namespace(config_file='configs\7359-frackles.yaml', opts=[])
2020-07-06 19:11:33,163 logger INFO: World size: 1
2020-07-06 19:11:33,163 logger INFO: Loaded configuration file configs\7359-frackles.yaml
2020-07-06 19:11:33,163 logger INFO:
NAME: 7359-frackles-test
PPL_CELEBA_ADJUSTMENT: True
DATASET:
PART_COUNT: 16
SIZE: 20000
SIZE_TEST: 49000-20000
PATH: M:/dev/ALAE/project/ALAE-master/data/datasets/frackleLeft_20200108_x128color-dataset/frackleLeft_20200108_x128-dataset-r%02d.tfrecords.%03d
PATH_TEST: M:/dev/ALAE/project/ALAE-master/data/datasets/frackleLeft_20200108_x128color-dataset/frackleLeft_20200108_x128-dataset-r%02d.tfrecords.%03d
MAX_RESOLUTION_LEVEL: 7
STYLE_MIX_PATH: style_mixing/test_images/set_celeba
MODEL:
LATENT_SPACE_SIZE: 256
LAYER_COUNT: 6
MAX_CHANNEL_COUNT: 256
START_CHANNEL_COUNT: 64
DLATENT_AVG_BETA: 0.995
MAPPING_LAYERS: 8
OUTPUT_DIR: training_artifacts/7359-frackles-test
TRAIN:
BASE_LEARNING_RATE: 0.002
EPOCHS_PER_LOD: 6
LEARNING_DECAY_RATE: 0.1
LEARNING_DECAY_STEPS: []
TRAIN_EPOCHS: 80
LOD_2_BATCH_8GPU: [512, 256, 128, 64, 32, 32, 32, 32, 32]
LOD_2_BATCH_4GPU: [512, 256, 128, 64, 32, 32, 32, 32, 16]
LOD_2_BATCH_2GPU: [128, 128, 128, 64, 32, 32, 16]
LOD_2_BATCH_1GPU: [128, 128, 128, 64, 32, 16]
LEARNING_RATES: [0.0015, 0.0015, 0.0015, 0.0015, 0.0015, 0.0015, 0.002, 0.003, 0.003]
2020-07-06 19:11:33,164 logger INFO: Running with config:
DATASET:
FFHQ_SOURCE: /data/datasets/ffhq-dataset/tfrecords/ffhq/ffhq-r%02d.tfrecords
FLIP_IMAGES: True
MAX_RESOLUTION_LEVEL: 7
PART_COUNT: 16
PART_COUNT_TEST: 1
PATH: M:/dev/ALAE/project/ALAE-master/data/datasets/frackleLeft_20200108_x128color-dataset/frackleLeft_20200108_x128-dataset-r%02d.tfrecords.%03d
PATH_TEST: M:/dev/ALAE/project/ALAE-master/data/datasets/frackleLeft_20200108_x128color-dataset/frackleLeft_20200108_x128-dataset-r%02d.tfrecords.%03d
SAMPLES_PATH: dataset_samples/faces/realign128x128
SIZE: 20000
SIZE_TEST: 29000
STYLE_MIX_PATH: style_mixing/test_images/set_celeba
MODEL:
CHANNELS: 3
DLATENT_AVG_BETA: 0.995
ENCODER: EncoderDefault
GENERATOR: GeneratorDefault
LATENT_SPACE_SIZE: 256
LAYER_COUNT: 6
MAPPING_FROM_LATENT: MappingFromLatent
MAPPING_LAYERS: 8
MAPPING_TO_LATENT: MappingToLatent
MAX_CHANNEL_COUNT: 256
START_CHANNEL_COUNT: 64
STYLE_MIXING_PROB: 0.9
TRUNCATIOM_CUTOFF: 8
TRUNCATIOM_PSI: 0.7
Z_REGRESSION: False
NAME: 7359-frackles-test
OUTPUT_DIR: training_artifacts/7359-frackles-test
PPL_CELEBA_ADJUSTMENT: True
TRAIN:
ADAM_BETA_0: 0.0
ADAM_BETA_1: 0.99
BASE_LEARNING_RATE: 0.002
EPOCHS_PER_LOD: 6
LEARNING_DECAY_RATE: 0.1
LEARNING_DECAY_STEPS: []
LEARNING_RATES: [0.0015, 0.0015, 0.0015, 0.0015, 0.0015, 0.0015, 0.002, 0.003, 0.003]
LOD_2_BATCH_1GPU: [128, 128, 128, 64, 32, 16]
LOD_2_BATCH_2GPU: [128, 128, 128, 64, 32, 32, 16]
LOD_2_BATCH_4GPU: [512, 256, 128, 64, 32, 32, 32, 32, 16]
LOD_2_BATCH_8GPU: [512, 256, 128, 64, 32, 32, 32, 32, 32]
REPORT_FREQ: [100, 80, 60, 30, 20, 10, 10, 5, 5]
SNAPSHOT_FREQ: [300, 300, 300, 100, 50, 30, 20, 20, 10]
TRAIN_EPOCHS: 80
Running on GeForce RTX 2080 Ti
2020-07-06 19:11:35,057 logger INFO: Trainable parameters generator:
2020-07-06 19:11:35,059 logger INFO: Trainable parameters discriminator:
2020-07-06 19:11:35,062 logger INFO: No checkpoint found. Initializing model from scratch
2020-07-06 19:11:35,062 logger INFO: Starting from epoch: 0
2020-07-06 19:11:35,116 logger INFO: ################################################################################
2020-07-06 19:11:35,117 logger INFO: # Switching LOD to 0
2020-07-06 19:11:35,117 logger INFO: # Starting transition
2020-07-06 19:11:35,117 logger INFO: ################################################################################
2020-07-06 19:11:35,117 logger INFO: ################################################################################
2020-07-06 19:11:35,117 logger INFO: # Transition ended
2020-07-06 19:11:35,117 logger INFO: ################################################################################
2020-07-06 19:11:35,119 logger INFO: Batch size: 128, Batch size per GPU: 128, LOD: 0 - 4x4, blend: 1.000, dataset size: 20000
Backend TkAgg is interactive backend. Turning interactive mode on._
Process finished with exit code -1073741819 (0xC0000005)
When debugging in PyCharm I have found that the error occures on line 74 in data_loader.py when calling b = next(yielder)
But since I have a very little experience in debugging python I would be glad if you know what might me a problem.
Thank you very much in advance.
Hello authors,
Thank you for your incredibly interesting paper. I had a very quick question about the CelebA-HQ train/test split.
I believe the config uses the split 29000/1000: https://github.com/podgorskiy/ALAE/blob/master/configs/celeba-hq256.yaml#L6-L7
And in the paper (page 7, bottom left column), you say: "We follow [16, 17, 27, 23] and use CelebAHQ downscaled to 256 ร 256 with training/testing split of 27000/3000."
If I am looking to compare results, which split size should I use -- or is there something here that I am missing?
Thank you!
Hi,
I see that you have implemented TFrecords for ImageNet in ALAE/dataset_preparation/prepare_imagenet.py
. Do you also have trained models for this dataset, and if yes what was your experience ?
Thanks for the information!
i can't pip DareBlopy in windows,how to repair it ,thank u
Hello, I'm trying to get this running in a conda environment and running into SIGABRT when starting training.
I've installed all the listed dependencies and created the tfrecords for my dataset by modifying prepare_celeba_hq_tfrecords.py to grab images from my own folder instead. This all seemed to go fine, but when training I get the following error:
Traceback (most recent call last):
File "train_alae.py", line 375, in <module>
run(train, get_cfg_defaults(), description="StyleGAN", default_config="configs/ffhq.yaml", world_size=gpu_count)
File "/home/hans/code/ALAE/launcher.py", line 122, in run
mp.spawn(_run, args=(world_size, fn, defaults, write_log, no_cuda, args), nprocs=world_size, join=True)
File "/home/hans/.conda/envs/alae/lib/python3.6/site-packages/torch/multiprocessing/spawn.py", line 200, in spawn
return start_processes(fn, args, nprocs, join, daemon, start_method='spawn')
File "/home/hans/.conda/envs/alae/lib/python3.6/site-packages/torch/multiprocessing/spawn.py", line 158, in start_processes
while not context.join():
File "/home/hans/.conda/envs/alae/lib/python3.6/site-packages/torch/multiprocessing/spawn.py", line 108, in join
(error_index, name)
Exception: process 0 terminated with signal SIGABRT
The traceback isn't very useful, but I believe it's happening on the model.train()
call just before for x_orig in tqdm(batches):
in train_alae.py
.
I'm running this on Ubuntu 18.10 with 2x1080Ti GPUs with NVidia driver version: 435.21 / CUDA version: 10.1.
I've tried installing cudatoolkit=9.0 with conda, but then the environment can only solve up to pytorch=1.1. With cudatoolkit=9.2 I was able to get pytorch=1.3, but with the same result.
My conda env:
# packages in environment at /home/hans/.conda/envs/alae:
#
# Name Version Build Channel
_libgcc_mutex 0.1 main
absl-py 0.9.0 pypi_0 pypi
astor 0.8.1 pypi_0 pypi
blas 1.0 mkl
ca-certificates 2020.1.1 0
certifi 2020.4.5.1 py36_0
cudatoolkit 10.1.243 h6bb024c_0
cycler 0.10.0 pypi_0 pypi
dareblopy 0.0.2 pypi_0 pypi
dlutils 0.0.12 pypi_0 pypi
dnnlib 0.0.1 pypi_0 pypi
freetype 2.9.1 h8a8886c_1
future 0.18.2 pypi_0 pypi
gast 0.3.3 pypi_0 pypi
grpcio 1.28.1 pypi_0 pypi
imageio 2.8.0 pypi_0 pypi
intel-openmp 2020.0 166
joblib 0.14.1 pypi_0 pypi
jpeg 9b h024ee3a_2
kiwisolver 1.2.0 pypi_0 pypi
ld_impl_linux-64 2.33.1 h53a641e_7
libedit 3.1.20181209 hc058e9b_0
libffi 3.2.1 hd88cf55_4
libgcc-ng 9.1.0 hdf63c60_0
libgfortran-ng 7.3.0 hdf63c60_0
libpng 1.6.37 hbc83047_0
libstdcxx-ng 9.1.0 hdf63c60_0
libtiff 4.1.0 h2733197_0
markdown 3.2.1 pypi_0 pypi
matplotlib 3.2.1 pypi_0 pypi
mkl 2020.0 166
mkl-service 2.3.0 py36he904b0f_0
mkl_fft 1.0.15 py36ha843d7b_0
mkl_random 1.1.0 py36hd6b4f25_0
ncurses 6.2 he6710b0_0
ninja 1.9.0 py36hfd86e86_0
numpy 1.14.5 pypi_0 pypi
olefile 0.46 py36_0
openssl 1.1.1g h7b6447c_0
packaging 20.3 pypi_0 pypi
pillow 7.0.0 py36hb39fc2d_0
pip 20.0.2 py36_1
protobuf 3.11.3 pypi_0 pypi
pyparsing 2.4.7 pypi_0 pypi
python 3.6.10 hcf32534_1
python-dateutil 2.8.1 pypi_0 pypi
pytorch 1.5.0 py3.6_cuda10.1.243_cudnn7.6.3_0 pytorch
pyyaml 5.3.1 pypi_0 pypi
readline 8.0 h7b6447c_0
scikit-learn 0.22.2.post1 pypi_0 pypi
scipy 1.4.1 pypi_0 pypi
setuptools 39.1.0 pypi_0 pypi
six 1.14.0 py36_0
sklearn 0.0 pypi_0 pypi
sqlite 3.31.1 h62c20be_1
tensorboard 1.10.0 pypi_0 pypi
tensorflow-gpu 1.10.0 pypi_0 pypi
termcolor 1.1.0 pypi_0 pypi
tk 8.6.8 hbc83047_0
torchvision 0.6.0 py36_cu101 pytorch
tqdm 4.45.0 pypi_0 pypi
werkzeug 1.0.1 pypi_0 pypi
wheel 0.34.2 py36_0
xz 5.2.5 h7b6447c_0
yacs 0.1.7 pypi_0 pypi
zlib 1.2.11 h7b6447c_3
zstd 1.3.7 h0b5b093_0
0
I tested interactive_demo.py on p2.x8large(8GPUs)
but I encountered
Segmentation fault (core dumped)
Could you give me advice which part should I fix for this problem??
Dear @podgorskiy :
Thank you for your great work ALAE! when I Continue to train the LOD=7 , I load checkpoint from lod_6.pth, but the loss is 100 times larger than LOD6 , and the sample image is bad . not a normal RGB image. It seems not load checkpoint well. When I sample images after each loss.backward ,I found the first picture(LOD=7, after the first loss.backward) is normal, as the time goes, the picture is broken dowm. It seems checkpoint is load well, but somthine about LOD6 and LOD7 is lost, Counld you help me ? Thank you !
I used pip install dareblopy to reinstall dareblopy. However, after reinstallment, the problem still exists, and I make sure the path to tfrecords I created by python dataset_preparation/prepare_celeba_hq_tfrecords.py is correct.
When I use this code https://github.com/VSehwag/ALAE/blob/master/replicate_results.ipynb
I get this results
Why?
P.S. I plot only one image
UPD: probably there were error in my files
In https://github.com/podgorskiy/ALAE/blob/master/train_alae.py#L293 we are setting the input tensor to require gradients. Why is this? Shouldn't the input data tensor be untouched by optimization?
Hey, it's a super interesting paper and reading experience of the same was awesome. Really interesting work.
I am trying to replicate your work to get more insight. I tried training on Google CoLab, Here's the link to the notebook: https://colab.research.google.com/drive/14CpH6eU4XsHPN_y4lhfpEZGHPpXT0AxL
I am not able to understand what's happening. In case I am doing something wrong do correct me. If I am not following the instructions properly do point me to the step I am missing.
Thanks in advance :D
A declarative, efficient, and flexible JavaScript library for building user interfaces.
๐ Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. ๐๐๐
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google โค๏ธ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.