def from_pretrained(self):
checkpoint_path = module_path + "/{}.pt".format(self.model_name)
if not os.path.exists(checkpoint_path):
checkpoint_url = f'https://drive.google.com/uc?id={model_urls[self.model_name]}'
gdown.cached_download(checkpoint_url, checkpoint_path)
self.model.load_state_dict(torch.load(checkpoint_path, map_location=self.device))
return self
def setup_repository():
# Downloading, extracting models.
models_url = 'https://drive.google.com/uc?id=1QJZWF9CzgOiYzjzsRSu2LOkrzi2S6j_U'
models_path = os.path.join(os.path.dirname(os.path.realpath(__file__)),
'resources', 'models')
os.makedirs(models_path, exist_ok=True)
md5 = '8920cc50fee3505e958307fa11088c0d'
models_archive_path = os.path.join(
os.path.dirname(os.path.realpath(__file__)), 'models.zip')
gdown.cached_download(url=models_url, path=models_archive_path, md5=md5)
gdown.extractall(path=models_archive_path, to=models_path)
os.remove(models_archive_path)
# Setting up the data folder with runtime_config.ini file
data_path = os.path.join(os.path.dirname(os.path.realpath(__file__)),
'resources', 'data')
os.makedirs(data_path, exist_ok=True)
runtime_config_path = os.path.join(data_path, 'runtime_config.ini')
if os.path.exists(runtime_config_path):
os.remove(runtime_config_path)
pfile = open(runtime_config_path, 'w')
pfile.write("[Predictions]\n")
pfile.write("non_overlapping=true\n")
pfile.write(
"reconstruction_method=probabilities #probabilities, thresholding\n")
pfile.write(
"reconstruction_order=resample_first #resample_first, resample_second\n"
)
pfile.write("probability_threshold=0.4\n")
pfile.close()
def download_and_load_model(model_files) -> RecursiveScriptModule:
"""
Downloads and torch.jit.load the model from google drive, the downloaded model is saved in /tmp
since in heroku we get /tmp to save all our stuff, if the app is not running in production
the model must be saved in load storage, hence the model is directly loaded
Args:
model_files: the dict containing the model information
Returns:
(RecursiveScriptModule): the loaded torch.jit model
"""
if "PRODUCTION" in os.environ:
logger.info(
f"=> Downloading Model {model_files['model_file']} from {model_files['model_url']}"
)
# heroku gives you `/tmp` to store files, which can be cached
model_path: Path = Path("/tmp") / f"{model_files['model_file']}.pt"
if not model_path.exists():
gdown.cached_download(url=model_files["model_url"], path=model_path)
logger.info(f"=> Loading {model_files['model_file']} from download_cache")
model: RecursiveScriptModule = torch.jit.load(str(model_path))
else:
logger.info(f"=> Loading {model_files['model_file']} from Local")
model = torch.jit.load(
str((Path("models") / (model_files["model_file"] + ".pt")))
)
return model
def creator(path):
train_path = gdown.cached_download(train_url)
dev_path = gdown.cached_download(dev_url)
test_path = gdown.cached_download(test_url)
dataset = {}
for split in ("train", "dev", "test"):
data_path = {
"train": train_path,
"dev": dev_path,
"test": test_path
}[split]
with io.open(data_path, "rt", encoding="utf-8") as f:
data = [json.loads(line) for line in f.readlines()]
temp = []
for x in data:
answer_key = x["answerKey"] if split != "test" else ""
options = {
choice["label"]: choice["text"]
for choice in x["question"]["choices"]
}
stem = x["question"]["stem"]
temp.append({
"id": x["id"],
"answer_key": answer_key,
"options": options,
"stem": stem
})
dataset[split] = temp
with io.open(path, "wb") as f:
pickle.dump(dataset, f)
return dataset
def setup_repository():
# Downloading, extracting models.
models_url = 'https://drive.google.com/uc?id=1DBIl8JyXEo6YdM9uNyo3vrv5T2WsYSXT'
models_path = os.path.join(os.path.dirname(os.path.realpath(__file__)),
'resources', 'models')
os.makedirs(models_path, exist_ok=True)
md5 = '434775bebd64910e01f4198eab251666'
models_archive_path = os.path.join(
os.path.dirname(os.path.realpath(__file__)), 'models.zip')
gdown.cached_download(url=models_url, path=models_archive_path, md5=md5)
gdown.extractall(path=models_archive_path, to=models_path)
os.remove(models_archive_path)
# Setting up the data folder with runtime_config.ini file
data_path = os.path.join(os.path.dirname(os.path.realpath(__file__)),
'resources', 'data')
os.makedirs(data_path, exist_ok=True)
runtime_config_path = os.path.join(data_path, 'runtime_config.ini')
if os.path.exists(runtime_config_path):
os.remove(runtime_config_path)
pfile = open(runtime_config_path, 'w')
pfile.write("[Predictions]\n")
pfile.write("non_overlapping=true\n")
pfile.write(
"reconstruction_method=probabilities #probabilities, thresholding\n")
pfile.write(
"reconstruction_order=resample_first #resample_first, resample_second\n"
)
pfile.write("probability_threshold=0.4\n")
pfile.close()
def get_bag_file(bag_id):
if bag_id == "dynamic.desk":
bag_file = gdown.cached_download(
url=
"https://drive.google.com/uc?id=1vaPyJERDNFY7W8VUBT3JvPNXHpZ6Bqun", # NOQA
md5="5bfb6eb7f80773dd2b8a6b16a93823e9",
)
elif bag_id == "static.robot":
bag_file = gdown.cached_download(
url=
"https://drive.google.com/uc?id=1DwSQLhrtciKCO566R2TqdPc81QOi7tlN", # NOQA
md5="ec155c5eacab3d90e72d7b1286c7aff3",
)
elif bag_id == "static.desk":
bag_file = gdown.cached_download(
url=
"https://drive.google.com/uc?id=1mArVTWl2f0Uws_mRoCtDzttV2s6AahZa", # NOQA
md5="a70a792577447a414c8ac7fe5f4aa316",
)
elif bag_id == "static.desk.topdown":
bag_file = gdown.cached_download(
url=
"https://drive.google.com/uc?id=1UxUg4IozQvQNrCALXkzk23v3fBjpFCqZ", # NOQA
md5="3625c11cd130a06557f38b8ff390882e",
)
else:
raise ValueError(f"Unknown bag_id: {bag_id}")
return bag_file
def model_init_app(reset=False):
model_path = current_app.config["MODEL_DIRECTORY"]
# check if model path exists
if not os.path.exists(model_path):
# check if parent directory exists
parent_dir = os.path.dirname(model_path)
if not os.path.exists(parent_dir):
os.mkdir(parent_dir)
current_app.logger.info(f"Created parent dir {parent_dir}")
# download and unzip to correct path
url = 'https://drive.google.com/uc?id=1XwjMGcYIg2qwEKHsC7uSmZayHvnEFhyg'
output = os.path.join(parent_dir, 'warpgan_pretrained.zip')
current_app.logger.debug("Downloading pretrained model")
gdown.cached_download(url,
output,
quiet=False,
postprocess=gdown.extractall)
current_app.logger.debug("Download complete")
os.remove(output) # remove downloaded zip after extraction
else:
current_app.logger.info("Pretrained model found - loading local file")
if reset:
pass
network = WarpGAN()
network.load_model(model_path)
return network
def download(self):
gdown.cached_download(
url=
"https://drive.google.com/uc?id=1l0ki7dX1WxcmV5Tfm41FPW-yk-wKUfne", # NOQA
path=self.root_dir + ".zip",
postprocess=gdown.extractall,
)
def load_dataset():
"""Load and return the dataset
Returns
-------
data : TODO
--------
"""
destination = os.path.join(DATA_DIR, 'dataset.jsons.gz')
gdown.cached_download(dataset_url,
destination,
md5='a8f860b2dc400e14d4e7083775c1308a',
quiet=False)
if not os.path.isfile(destination):
# Data not cached, download
urllib.request.urlretrieve(dataset_url, destination)
source_file = tf.constant([destination])
dataset = tf.data.TextLineDataset(filenames=source_file,
compression_type='GZIP')
dataset = dataset.map(
lambda x: tf.py_function(_serialize, [x], [tf.string]))
dataset = dataset.map(_parse)
return dataset
def pr2_urdfpath():
gdown.cached_download(
url='https://drive.google.com/uc?id=1zy4C665o6efPko7eMk4XBdHbvgFfdC-6',
path=osp.join(download_dir, 'pr2_description.tar.gz'),
md5='e4fb915accdb3568a5524c92e9c35c9a',
postprocess=gdown.extractall,
)
return osp.join(download_dir, 'pr2_description/pr2.urdf')
def panda_urdfpath():
gdown.cached_download(
url='https://drive.google.com/uc?id=1h6ib9jpEUNa1xB2DNrnRQtqpSD2Rj9bz',
path=osp.join(download_dir, 'franka_description.tar.gz'),
md5='3de5bd15262b519e3beb88f1422032ac',
postprocess=gdown.extractall,
)
return osp.join(download_dir, 'franka_description/panda.urdf')
def main():
# __ https://drive.google.com/file/d/1zo4BkS8wqqbc7gxYQMedhqCiQUb51LAp/view?usp=sharing
url = "https://drive.google.com/uc?id=0B9P1L--7Wd2vNm9zMTJWOGxobkU"
output = "20150428_collected_images.tgz"
gdown.download(url, output, quiet=False)
md5 = "fa837a88f0c40c513d975104edf3da17"
gdown.cached_download(url, output, md5=md5, postprocess=gdown.extractall)
gdown.download(url)
def contrib_file(download_dir):
"""Download the contributon file
"""
contrib_file = download_dir / 'contrib.deeplift.h5'
url = 'https://drive.google.com/uc?id=1-70VlFvcOCwwt4SrEXoqkaXyBQPnlQGZ'
md5 = '56e456f0d1aeffc9d3fcdfead0520c17'
gdown.cached_download(url, str(contrib_file), md5=md5)
return contrib_file
def fetch_urdfpath():
gdown.cached_download(
url='https://drive.google.com/uc?id=1y7Jc3QoVW6J072CrSNupfKpyLp4NNxuH',
path=osp.join(download_dir, 'fetch_description.tar.gz'),
md5='fbe29ab5f3d029d165a625175b43a265',
postprocess=gdown.extractall,
)
return osp.join(download_dir, 'fetch_description/fetch.urdf')
def gdrive_download(url, folder, filename):
cache_dir = os.path.join(os.getenv("HOME"), ".cache", "torchfly", folder)
os.makedirs(cache_dir, exist_ok=True)
filepath = os.path.join(cache_dir, filename)
gdown.cached_download(url, filepath, quiet=False)
return filepath
def download(url, path=None, cached=True):
import gdown
# kitti_url = 'https://drive.google.com/uc?id=1QHvE8oHlHqXB97RHlulLuWCmleE0wZ8B'
# kitti_out = 'kitti.tar.bz2'
gdown.cached_download(url,
path=path,
quiet=False,
proxy=False,
postprocess=gdown.extractall)
def download(cls) -> None:
url: str = "https://drive.google.com/uc?id=1if4VoEXNx9W3XCn0Y7Fp15B4GpcYbyYi" # NOQA
md5 = None # 'c9122e177a766a9691cab13c5cda41a9'
gdown.cached_download(
url=url,
path=cls._root_dir + ".zip",
md5=md5,
postprocess=gdown.extractall,
)
def bunny_objpath():
target_path = osp.join(download_dir, 'mesh', 'bunny.obj')
gdown.cached_download(
url='https://drive.google.com/uc?id=18aAYzBglAGaSwes6oAENcufD8KC5XDph',
path=target_path,
md5='19bd31bde1fcf5242a8a82ed4ac03c72',
quiet=True,
)
return target_path
def download(self):
logger.info(
f"=> Downloading GDrive file {self.file_name} from {self.gdrive_url}"
)
file_path: Path = get_temp_folder() / f"{self.file_name}"
if not file_path.exists():
gdown.cached_download(url=self.gdrive_url, path=file_path)
return file_path
def download_ckpt(data_dir, name, download_info):
if os.path.exists(join(data_dir, 'checkpoints', name)):
print('\nINFO: {} model already downloaded.'.format(name))
else:
print('\nINFO: Downloading {} model...'.format(name))
url = 'https://drive.google.com/uc?id=' + download_info['url']
zip_output = join(data_dir, 'checkpoints', name + '.zip')
md5 = download_info['md5']
gdown.cached_download(url, zip_output, md5=md5, postprocess=gdown.extractall, quiet=True)
os.remove(zip_output)
def _download_unity(self):
url = "https://drive.google.com/uc?id=" + APP_GDRIVE_ID[platform]
os.makedirs("binary", exist_ok=True)
# zip_path = os.path.join("binary", APP_FILE_NAME[platform])
zip_path = APP_FILE_NAME[platform]
if os.path.exists(zip_path):
logger.info("%s is already downloaded.", zip_path)
else:
logger.info("Downloading Unity app from %s", url)
gdown.cached_download(url, zip_path, postprocess=gdown.extractall)
def download(cls) -> None:
url: str = "https://drive.google.com/uc?id=1PKEJ8SVIVLukvmeIBexp6_XmJetHoOf2" # NOQA
md5: str = "540c37435e4a16546850a83690b2db9b"
gdown.cached_download(
url=url,
path=cls._root_dir + ".zip",
md5=md5,
postprocess=gdown.extractall,
)
def _SSD_VGG_VOC_weights():
import gdown
url = 'https://drive.google.com/uc?id=1GP6i9lcEqK3mOgr7soxruLhWpITxShge'
output = Path.home() / '.anchors-jax/ssd-coco.jax'
output.parent.mkdir(exist_ok=True, parents=True)
gdown.cached_download(url, str(output))
params = pickle.load(output.open('rb'))
return _split_bb_ssd_params(params)
def get_path(cls, quiet=True):
"""
returns the mnist folder, downloading the mnist data if needed
"""
if cls._path is None:
url = 'https://drive.google.com/uc?id=1im9_agM-hHl8dKQU5uWh054rScjPINtz' # jeff hykin's google drive copy of mnist
output = f'{temp_folder_path}/mnist.zip'
gdown.cached_download(url, output, postprocess=gdown.extractall, quiet=quiet)
cls._path = temp_folder_path+"/mnist"
return cls._path
def creator(path):
dataset = {}
for split in ('train', 'dev', 'test'):
en_path = gdown.cached_download(en_url.format(split))
ja_path = gdown.cached_download(ja_url.format(split))
with io.open(en_path, 'rt') as en, io.open(ja_path, 'rt') as ja:
dataset[split] = [(x.rstrip(os.linesep), y.rstrip(os.linesep))
for x, y in zip(en, ja)]
with io.open(path, 'wb') as f:
pickle.dump(dataset, f)
return dataset
def setup_repository():
# Downloading, extracting models.
models_url = 'https://drive.google.com/uc?id=1ga08d8QQfAHOgTSKiPpIN7f_owuicNUA'
models_path = os.path.join(os.path.dirname(os.path.realpath(__file__)),
'resources', 'models')
os.makedirs(models_path, exist_ok=True)
md5 = '55c66e000de9077e483635029f740901'
models_archive_path = os.path.join(
os.path.dirname(os.path.realpath(__file__)), 'models.zip')
gdown.cached_download(url=models_url, path=models_archive_path, md5=md5)
gdown.extractall(path=models_archive_path, to=models_path)
os.remove(models_archive_path)
def __init__(self):
root_dir = morefusion.utils.get_data_path(
"wkentaro/morefusion/ycb_video/real_data/20191212_163242.566559922"
)
gdown.cached_download(
url=
"https://drive.google.com/uc?id=1llWN7MOLzJZnaRDD4XGSmRWAFBtP3P9z", # NOQA
md5="a773bb947377811b2b66ab9bc17f4d8d",
path=root_dir + ".zip",
)
self._dataset = morefusion.datasets.MyRealRGBDPoseEstimationDataset(
root_dir=root_dir, )
def prepare_data(self):
if os.path.exists(join(self.root_dir,
self.DIR_NAME)) and os.path.isdir(
join(self.root_dir, self.DIR_NAME)):
if get_dir_md5hash(join(self.root_dir,
self.DIR_NAME)) == self.DIR_HASH:
return True
raise ValueError('Wrong checksum, delete %s dir' % self.root_dir)
gdown.cached_download(
'https://drive.google.com/uc?id=1-yZki-hyVcHKWB4VfqAUaAfBazPyemqN',
os.path.join(self.root_dir, 'acdc.tar.gz'),
md5=self.TAR_HASH,
postprocess=gdown.extractall)
return True
def _download_dataset(self) -> None:
base_url = self.module.base_gdrive_url
all_name_url_md5 = [(item[0], base_url + item[1], item[2])
for item in self.module.name_gdriveid_md5]
for name, gdrive_url, md5 in all_name_url_md5:
if self.verbose:
print("Downloading " + name + "...")
filepath = self.root / name
gdown.download(gdrive_url, str(filepath), quiet=False)
gdown.cached_download(gdrive_url, str(filepath), md5=md5)
self._extract_archive(filepath, remove_archive=True)
def load_state_dict_from_url(url, name):
# moslty code from torch.hub
torch_home = _get_torch_home()
model_dir = os.path.join(torch_home, "checkpoints")
os.makedirs(model_dir, exist_ok=True)
cached_file = os.path.join(model_dir, f"{name}.pth")
gdown.cached_download(url, path=cached_file, quiet=False)
state_dict = torch.load(cached_file)["state_dict"]
state_dict = OrderedDict(
{key[7:]: value
for key, value in state_dict.items()})
return state_dict
