def __init__(self, opt):
"""
Core Atlasnet module : decoder to meshes and pointclouds.
This network takes an embedding in the form of a latent vector and returns a pointcloud or a mesh
Author : Thibault Groueix 01.11.2019
:param opt:
"""
super(Atlasnet, self).__init__()
self.opt = opt
self.device = opt.device
# Define number of points per primitives
self.nb_pts_in_primitive = opt.number_points // opt.nb_primitives
self.nb_pts_in_primitive_eval = opt.number_points_eval // opt.nb_primitives
if opt.remove_all_batchNorms:
torch.nn.BatchNorm1d = Identity
print("Replacing all batchnorms by identities.")
# Initialize templates
self.template = [
get_template(opt.template_type, device=opt.device)
for i in range(0, opt.nb_primitives)
]
# Intialize deformation networks
self.decoder = nn.ModuleList(
[Mapping2Dto3D(opt) for i in range(0, opt.nb_primitives)])
def get():
id = request.args.get('id')
if not (id and ObjectId.is_valid(id)):
abort(400)
result = get_template(ObjectId(id))
if not result:
return Response(status=404)
obj = convert_template_from_mongo(result)
response_data = json.dumps(obj)
logger.debug('--response_data--' + response_data)
resp = Response(response=response_data, status=201, content_type='application/json')
return resp
def prepare_image_regions(id):
image = get_image(id)
if not image:
return False
kind = image.get('kind')
if not kind:
return False
storage_id = image.get('storage_id')
file = read_gridfs(storage_id)
if not file:
return False
filename = file.filename
bytes = file.read()
template_id = image.get('template_id')
template = get_template(template_id)
if not template:
return False
regions = template.get('regions')
if not regions:
return False
path = os.path.join(image_dir, kind, str(id))
if not os.path.exists(path):
os.makedirs(path)
src_filename = os.path.join(path, filename)
logger.debug('--src_filename--' + src_filename)
f = open(src_filename, 'wb')
f.write(bytes)
f.close()
try:
image_collection.update_one({'_id': id}, {
'$set': {
'filename': filename,
'regions': regions,
'status': 'prepare'
}
})
except Exception as e:
logger.debug('--prepare_image_regions--' + str(e))
return False
return True
def __init__(self, opt, proto_feature, ttl_point, num_primitives):
super(Proto2Dto3DCluster, self).__init__()
self.opt = opt
self.proto_feat = nn.Parameter(
torch.from_numpy(proto_feature).float().unsqueeze(0),
requires_grad=False) # Dummy var for initialization
self.num_slaves = num_primitives
self.num_per_slave = ttl_point // num_primitives
self.template = [
get_template(opt.template_type, device=opt.device)
for _ in range(num_primitives)
]
print(
f'New 2D -> 3D Cluster with prototype initialized, this cluster contains {self.num_slaves} slave(s)'
)
self.slave_pool = nn.ModuleList(
[Proto2Dto3DSlave(opt) for _ in range(num_primitives)])