Skip to content
Snippets Groups Projects
Commit 9382dc5a authored by Luc Giffon's avatar Luc Giffon
Browse files

remove VGG19Transformer folder with irelevant stuf inside (Imagenet... 

remove VGG19Transformer folder with irelevant stuf inside (Imagenet Transformer and CovAbs transfmer)
parent aed430d4
Branches
No related tags found
No related merge requests found
Show whitespace changes
Inline Side-by-side
skluc/main/data/transformation/VGG19Transformer.py 0 → 100644
+ 0
− 0
View file @ 9382dc5a
skluc/main/data/transformation/VGG19Transformer/VGG19Cifar10CovAbs.py deleted 100644 → 0
+ 0
− 98
View file @ aed430d4
import os
import numpy as np
from keras import Model
from keras.models import load_model
from skluc.main.data import VGG19Transformer
from skluc.main.utils import logger, create_directory, download_data, check_file_md5, deprecated
# todo check those deprecated things
@deprecated
class VGG19Cifar10CovAbs(VGG19Transformer):
"""
Extend the vgg19transformer class with weights learned on CIFAR10.
The covariance matrix is then computed on the transformed data image.
"""
NAME = "vgg19_cifar10_cov"
MODEL_URL = "https://pageperso.lis-lab.fr/~luc.giffon/models/1522967518.1916964_vgg19_cifar10.h5"
MODEL_CHECKSUM = "0dbb4f02ceb1f4acb6e24831758106e5"
# todo faire une fonction qui regarde directement le checksum sur le site ?
def __init__(self):
super().__init__(name=self.NAME)
@staticmethod
def _compute_cov_matrix(data):
"""
:param data: (b x W x H x D)
:type data: np.ndarray
:return:
"""
data = data.reshape((data.shape[0], data.shape[1] * data.shape[2], data.shape[3]))
mean = np.mean(data, axis=1)
mean = mean.reshape((mean.shape[0], 1, mean.shape[-1]))
data_centered = data - mean
cov_mat = []
for i, mat in enumerate(data_centered):
cov_mat.append(mat.T.dot(mat))
if i % 1000 == 0:
logger.debug("Computing covariance matrix - step {}/{}".format(i, len(data_centered)))
cov_mat = 1. / data.shape[1] * np.array(cov_mat)
logger.debug("Final covariance matrix shape: {}".format(str(cov_mat.shape)))
return cov_mat
@staticmethod
def _compute_log_matrix(data):
log_mat = []
for i, mat in enumerate(data):
U, S, V = np.linalg.svd(mat, full_matrices=False)
log_mat.append(U.dot(np.diag(np.log(S))).dot(V))
if i % 1000 == 0:
logger.debug("Computing log matrix - step {}/{}".format(i, len(data)))
log_mat = np.array(log_mat)
logger.debug("Final log matrix shape: {}".format(str(log_mat.shape)))
return log_mat
def load(self):
create_directory(self.s_download_dir)
s_model_path = download_data(self.MODEL_URL, self.s_download_dir)
check_file_md5(s_model_path, self.__class__.MODEL_CHECKSUM)
if self.vgg_conv_model is None:
logger.debug("Loading VGG19 model with cifar10 weights")
self.vgg_conv_model = load_model(s_model_path)
bloc3pool_layer = self.vgg_conv_model.get_layer('block3_pool')
# this is weird but the index is actually the index of the layer just before the pooling layer
# so this is what we want here: we don't want the pooling
index_bloc3pool_layer = self.vgg_conv_model.layers.index(bloc3pool_layer)
self.vgg_conv_model = Model(inputs=self.vgg_conv_model.input,
outputs=self.vgg_conv_model.get_layer(index=index_bloc3pool_layer).output)
else:
logger.debug("Skip loading model VGG19 model with cifar10 weights. Already there.")
def transform(self, data, labels):
if len(data.shape) != 4:
raise AssertionError("Data shape should be of size 4 (image batch with channel dimension). "
"It is {}: {}. Maybe have you forgotten to reshape it to an image format?"
"".format(len(data.shape), data.shape))
self.load()
transformed_data, labels = super().transform(data, labels)
transformed_data = self._compute_cov_matrix(transformed_data)
return transformed_data, labels
def check_model(self):
name = os.path.basename(os.path.normpath(self.__class__.MODEL_URL))
s_file_path = os.path.join(self.s_download_dir, name)
if os.path.exists(s_file_path) and check_file_md5(s_file_path,
self.__class__.MODEL_CHECKSUM,
raise_=False):
return True
else:
return False
skluc/main/data/transformation/VGG19Transformer/VGG19ImageNetTransformer.py deleted 100644 → 0
+ 0
− 31
View file @ aed430d4
from keras.applications import VGG19
from skluc.main.data import VGG19Transformer
from skluc.main.utils import logger, deprecated, Singleton
@deprecated
class VGG19ImagenetTransformer(VGG19Transformer, metaclass=Singleton):
"""
Extend the vgg19transformer class with convolutional wieghts learned on the imagenet dataset.
"""
NAME = "vgg19_imagenet"
def __init__(self):
super().__init__(name=self.NAME)
def load(self, input_shape):
if self.vgg_conv_model is None:
logger.debug("Loading VGG19 model with imagenet weights from keras")
self.vgg_conv_model = VGG19(include_top=False, weights='imagenet', input_shape=input_shape)
else:
logger.debug("Skip loading model VGG19 model with imagenet weights. Already there.")
def transform(self, data, labels):
# todo trouver une solution pour ne pas avoir un copier collé entre cette classe et celle avec cifar
if len(data.shape) != 4:
raise AssertionError("Data shape should be of size 4 (image batch with channel dimension). "
"It is {}: {}. Maybe have you forgotten to reshape it to an image format?"
"".format(len(data.shape), data.shape))
self.load(input_shape=data[0].shape)
return super().transform(data, labels)
skluc/main/data/transformation/VGG19Transformer/__init__.py deleted 100644 → 0
+ 0
− 93
View file @ aed430d4
from keras import Model
from keras.models import load_model
from skluc.main.data.mldatasets.Cifar10Dataset import Cifar10Dataset
from skluc.main.data.transformation.KerasModelTransformer import KerasModelTransformer
from skluc.main.utils import logger, create_directory, download_data, check_file_md5, Singleton, DownloadableModel
class VGG19Transformer(KerasModelTransformer, metaclass=Singleton):
"""
Uses the vgg19 convolution network to transform data.
"""
MAP_DATA_MODEL = {
"svhn": DownloadableModel(
url="https://pageperso.lis-lab.fr/~luc.giffon/models/1529968150.5454917_vgg19_svhn.h5",
checksum="563a9ec2aad37459bd1ed0e329441b05"
),
"cifar100": DownloadableModel(
url="https://pageperso.lis-lab.fr/~luc.giffon/models/1530965727.781668_vgg19_cifar100fine.h5",
checksum="edf43e263fec05e2c013dd5a2128fc38"
),
"cifar10": DownloadableModel(
url="https://pageperso.lis-lab.fr/~luc.giffon/models/1522967518.1916964_vgg19_cifar10.h5",
checksum="0dbb4f02ceb1f4acb6e24831758106e5"
),
"siamese_omniglot_28x28": DownloadableModel(
url="https://pageperso.lis-lab.fr/~luc.giffon/models/1536244775.6502118_siamese_vgg19_omniglot_28x28_conv.h5",
checksum="90aec06e688ec3248ba89544a10c9f1f"
),
"omniglot_28x28": DownloadableModel(
url="https://pageperso.lis-lab.fr/~luc.giffon/models/1536764034.66037_vgg19_omniglot.h5",
checksum="ef1272e9c7ce070e8f70889ec58d1c33"
)
}
def __init__(self, data_name, cut_layer_name=None, cut_layer_index=None):
if data_name not in self.MAP_DATA_MODEL.keys():
raise ValueError("Unknown data name. Can't load weights")
if cut_layer_name is None and cut_layer_index is None:
logger.warning(
"Cut layer chosen automatically but it eventually will lead to an error in future: block5_pool should be specified explicitly")
cut_layer_name = "block5_pool"
if cut_layer_name is not None:
transformation_name = str(data_name) + "_" + self.__class__.__name__ + "_" + str(cut_layer_name)
elif cut_layer_index is not None:
transformation_name = str(data_name) + "_" + self.__class__.__name__ \
+ "_" + str(cut_layer_index)
# todo sauvegarder index / nom dans le meme dossier si c'est les meme
else:
raise AttributeError("Cut layer name or cut_layer index must be given to init VGG19Transformer.")
self.__cut_layer_name = cut_layer_name
self.__cut_layer_index = cut_layer_index
self.keras_model = None
super().__init__(data_name=data_name,
transformation_name=transformation_name)
def load(self):
create_directory(self.s_download_dir)
s_model_path = download_data(self.MAP_DATA_MODEL[self.data_name].url, self.s_download_dir)
check_file_md5(s_model_path, self.MAP_DATA_MODEL[self.data_name].checksum)
if self.keras_model is None:
logger.debug("Loading VGG19 model for {} transformation with {} weights".format(self.transformation_name, self.data_name))
self.keras_model = load_model(s_model_path)
logger.debug("Layers of model {}".format([l.name for l in self.keras_model.layers]))
if self.__cut_layer_index is not None:
cut_layer = self.keras_model.layers[-1]
self.__cut_layer_name = cut_layer.name
logger.debug(
"Found associated layer {} to layer index {}".format(self.__cut_layer_name, self.__cut_layer_index))
self.keras_model = Model(inputs=self.keras_model.input,
outputs=self.keras_model.get_layer(name=self.__cut_layer_name).output)
else:
logger.debug("Skip loading model VGG19 for {} transformation with {} weights. Already there.".format(
self.transformation_name,
self.data_name))
if __name__ == '__main__':
valsize = 10000
d = Cifar10Dataset(validation_size=valsize)
d.load()
d.to_image()
trans = VGG19Transformer(data_name="cifar10", cut_layer_name="block5_pool")
d.apply_transformer(transformer=trans)
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Please register or to comment