def __init__(self,mod_AE,ShapeData,NiterProjection,NiterGrad,GradType,OptimType,InterpFlag=False,periodicBnd=False,N_cov=0):
super(Model_4DVarNN_GradFP, self).__init__()
self.model_AE = mod_AE
with torch.no_grad():
print('Optim type %d'%OptimType)
self.OptimType = OptimType
#self.NRes = int(NiterResidual)
self.NRes = int(3)
self.NProjFP = int(NiterProjection)
self.NGrad = int(NiterGrad)
self.Ncov = N_cov
self.InterpFlag = InterpFlag
self.periodicBnd = periodicBnd
self.shape = ShapeData
# init median pooling
self.median_pooling = MedianPool2d()
# Define Solver type according to OptimType
## Gradient-based minimization using a fixed-step descent
if OptimType == 0:
self.model_Grad = model_GradUpdate0(replace_tup_at_idx(self.shape,0,int(self.shape[0]/(self.Ncov+1))),GradType)
## Gradient-based minimization using a CNN using a (sub)gradient as inputs
elif OptimType == 1:
self.model_Grad = model_GradUpdate1(replace_tup_at_idx(self.shape,0,int(self.shape[0]/(self.Ncov+1))),GradType,self.periodicBnd)
## Gradient-based minimization using a LSTM using a (sub)gradient as inputs
elif OptimType == 2:
self.model_Grad = model_GradUpdate2(replace_tup_at_idx(self.shape,0,int(self.shape[0]/(self.Ncov+1))),GradType,self.periodicBnd)
elif OptimType == 3:
self.model_Grad = model_GradUpdate2(replace_tup_at_idx(self.shape,0,int(self.shape[0]/(self.Ncov+1))),GradType,30)
def __init__(self):
super(PatchTransformer, self).__init__()
self.min_contrast = 0.8
self.max_contrast = 1.2
self.min_brightness = -0.1
self.max_brightness = 0.1
self.noise_factor = 0.10
self.minangle = -20 / 180 * math.pi
self.maxangle = 20 / 180 * math.pi
self.medianpooler = MedianPool2d(7, same=True)
'''
def __init__(self, config, width, height):
super(AdvPatch, self).__init__()
self.config = config
self.model_width = width
self.model_height = height
self.min_contrast = 0.8
self.max_contrast = 1.2
self.min_brightness = -0.1
self.max_brightness = 0.1
self.noise_factor = 0.10
self.medianpooler = MedianPool2d(3, same=True)
def __init__(self):
super(PatchTransformer_glasses, self).__init__()
self.min_contrast = 0.8
self.max_contrast = 1.2
self.min_brightness = -0.1
self.max_brightness = 0.1
self.noise_factor = 0.10
self.minangle = -10 / 180 * math.pi
self.maxangle = 10 / 180 * math.pi
self.medianpooler = MedianPool2d(
7, same=True) # kernel_size = 7? see again
def __init__(self):
super(PatchTransformer, self).__init__()
# os.environ['CUDA_VISIBLE_DEVICES'] = '3'
# self.device = 'cuda:3' if torch.cuda.is_available() else 'cpu'
self.min_contrast = 0.8
self.max_contrast = 1.2
self.min_brightness = -0.1
self.max_brightness = 0.1
self.noise_factor = 0.10
self.minangle = -20 / 180 * math.pi
self.maxangle = 20 / 180 * math.pi
self.medianpooler = MedianPool2d(
7, same=True) # kernel_size = 7? see again
'''
def __init__(self, config):
super(RenderModel, self).__init__()
self.config = config
self.min_contrast = 0.8
self.max_contrast = 1.2
self.min_brightness = -0.1
self.max_brightness = 0.1
self.noise_factor = 0.10
# if self.config.cuda is not '-1':
# torch.cuda.set_device(self.config.cuda)
# self.device = torch.device('cuda')
# else:
# self.device = torch.device('cpu')
# if self.config.consistent:
# self.grad_textures = grad_textutres.unsqueeze(-2).unsqueeze(-2).unsqueeze(-2)\
# .expand(self.config.depth * self.config.width * self.config.height, 4, 4, 4, 3)
# else:
# self.grad_textures = grad_textutres
# self.grad_textures = grad_textutres
# self.grad_textures = grad_textutres.expand(self.config.depth * self.config.width * self.config.height,
# 4, 4, 4, 3)
self.darknet_model = Darknet(self.config.cfgfile)
self.darknet_model.load_weights(self.config.weightfile)
self.darknet_model = self.darknet_model.eval().cuda()
# for p in self.darknet_model.parameters():
# p.requires_grad = False
# self.cubic = nn.Parameter(torch.full((1, 4, 4, 4, 3), 0.5).cuda())
# self.Xembedding = torch.nn.Embedding(100, 256)
# self.Yembedding = torch.nn.Embedding(100, 256)
# self.Zembedding = torch.nn.Embedding(100, 256)
# self.linear1 = nn.Linear(768, 192)
# self.linear2 = nn.Linear(256, 64)
# self.linear3 = nn.Linear(256, 64)
# self.linear4 = nn.Linear(192, 192)
# self.softmax = nn.Softmax(dim=2)
# self.convtranspose = nn.ConvTranspose3d(3, 3, (3, 3, 3), stride=1)
self.prob_extractor = MaxProbExtractor(0, 80).cuda()
self.nps_calculator = NPSCalculator(self.config.printfile,
self.config.image_size).cuda()
self.total_variation = TotalVariation().cuda()
self.medianpooler = MedianPool2d(7, same=True)
renderer = nr.Renderer(camera_mode='look_at')
renderer.perspective = False
renderer.light_intensity_directional = 0.0
renderer.light_intensity_ambient = 1.0
self.renderer = renderer