def forward(ctx, input, model, src, rec, device):
input = input.to('cpu')
# Save modeling parameters for backward pass
ctx.model = model
ctx.src = src
ctx.rec = rec
ctx.device = device
# Prepare input
input = input.detach()
model.dm = DevFunc(name="dm", grid=model.grid)
model.dm.data[:] = nn.ReflectionPad2d((40))(input).numpy()[0, 0, :, :]
# Linearized forward modeling
d_lin = forward_born(model, src.coordinates.data, src.data,
rec.coordinates.data)
return torch.from_numpy(d_lin).to(ctx.device)
src_type='Ricker',
f0=f0)
# Resample input data to computational grid
dorig = resample(dorig, t0, tn, nt, nt_comp)
dobs = Receiver(name='rec_t',
grid=model.grid,
ntime=nt_comp,
coordinates=rec_coordinates)
dobs.data[:] = dorig
# Predicted data
if iteration > 1:
dpred, summary0 = forward_born(model,
geometry.src_positions,
geometry.src.data,
geometry.rec_positions,
space_order=space_order)
# Residual and function value
dpred = dpred - dobs
else:
dpred = Receiver(name='rec',
grid=model.grid,
ntime=nt_comp,
coordinates=rec_coordinates)
dpred.data[:] = -dorig
# Function value
fval = np.array([.5 * np.linalg.norm(dpred.data)**2], dtype='float32')
print("fval, dpred.shape, dobs.shape: ", fval, dpred.shape, dobs.shape)
f0 = 0.010
src = RickerSource(name='src', grid=model.grid, f0=f0, time=time_axis)
src.coordinates.data[0, :] = np.array(model.domain_size) * 0.5
src.coordinates.data[0, -1] = 20.
# Receiver for observed data
rec_t = Receiver(name='rec_t', grid=model.grid, npoint=401, ntime=nt)
rec_t.coordinates.data[:, 0] = np.linspace(100, 900, num=401)
rec_t.coordinates.data[:, 1] = 20.
# Linearized data
print("Forward J")
t1 = time.time()
dD_hat = forward_born(model_const,
src.coordinates.data,
src.data,
rec_t.coordinates.data,
isic=False,
dt=dt)
dm_hat = model0.dm.data
t2 = time.time()
print(t2 - t1)
# Forward
print("Forward J")
t1 = time.time()
dD = forward_born(model0,
src.coordinates.data,
src.data,
rec_t.coordinates.data,
isic=True,
dt=dt)
t0=0.0,
tn=tn,
src_type='Ricker',
f0=f0)
# Resample input data to computational grid
dorig = resample(dorig, t0, tn, nt, nt_comp)
dobs = Receiver(name='rec_t',
grid=model.grid,
ntime=nt_comp,
coordinates=rec_coordinates)
dobs.data[:] = dorig
# Predicted data
if iteration > 1:
dpred = forward_born(model, geometry.src_positions, geometry.src.data,
geometry.rec_positions)[0]
# Residual and function value
dpred = dpred - dobs
else:
dpred = Receiver(name='rec',
grid=model.grid,
ntime=nt_comp,
coordinates=rec_coordinates)
dpred.data[:] = -dorig
# Function value
fval = np.array([.5 * np.linalg.norm(dpred.data)**2], dtype='float32')
print("fval, dpred.shape, dobs.shape: ", fval, dpred.shape, dobs.shape)
# Wavefields in memory
t1 = time.time()