def visualize_batch_np(self,
codes_all_arr,
truncation=None,
mean_latent=None,
B=15):
if truncation is None: truncation = self.truncation
if mean_latent is None: mean_latent = self.mean_latent
if self.StyleGAN.size == 1024: B = round(B / 4)
csr = 0
img_all = None
imgn = codes_all_arr.shape[0]
while csr < imgn:
csr_end = min(csr + B, imgn)
with torch.no_grad():
img_list = self.visualize(
torch.from_numpy(
codes_all_arr[csr:csr_end, :]).float().cuda(),
truncation=truncation,
mean_latent=mean_latent,
).cpu()
img_all = img_list if img_all is None else torch.cat(
(img_all, img_list), dim=0)
csr = csr_end
clear_output(wait=True)
progress_bar(csr_end, imgn,
"ploting row of page: %d of %d" % (csr_end, imgn))
return img_all
def get_full_hessian(loss_grad, model):
# from https://discuss.pytorch.org/t/compute-the-hessian-matrix-of-a-network/15270/3
cnt = 0
loss_grad = list(loss_grad)
for i, g in enumerate(loss_grad):
progress_bar(
i,
len(loss_grad),
"flattening to full gradient: %d of %d" % (i, len(loss_grad)),
)
g_vector = (g.contiguous().view(-1) if cnt == 0 else torch.cat(
[g_vector, g.contiguous().view(-1)]))
cnt = 1
hessian_size = g_vector.size(0)
hessian = torch.zeros(hessian_size, hessian_size)
for idx in range(hessian_size):
progress_bar(idx, hessian_size,
"full hessian columns: %d of %d" % (idx, hessian_size))
grad2rd = torch.autograd.grad(g_vector[idx],
model.parameters(),
create_graph=True)
cnt = 0
for g in grad2rd:
g2 = (g.contiguous().view(-1)
if cnt == 0 else torch.cat([g2, g.contiguous().view(-1)]))
cnt = 1
hessian[idx] = g2
return hessian.cpu().data.numpy()
def visualize_batch_np(self, codes_all_arr, truncation=0.7, B=5):
csr = 0
img_all = None
imgn = codes_all_arr.shape[0]
with torch.no_grad():
while csr < imgn:
csr_end = min(csr + B, imgn)
img_list = self.visualize(torch.from_numpy(codes_all_arr[csr:csr_end, :]).float().cuda(),
truncation=truncation, ).cpu()
img_all = img_list if img_all is None else torch.cat((img_all, img_list), dim=0)
csr = csr_end
clear_output(wait=True)
progress_bar(csr_end, imgn, "ploting row of page: %d of %d" % (csr_end, imgn))
return img_all
def power_iteration(
operator: Operator,
steps: int = 20,
error_threshold: float = 1e-4,
momentum: float = 0.0,
use_gpu: bool = True,
fp16: bool = False,
init_vec: torch.Tensor = None,
) -> Tuple[float, torch.Tensor]:
"""
Compute dominant eigenvalue/eigenvector of a matrix
operator: linear Operator giving us matrix-vector product access
steps: number of update steps to take
returns: (principal eigenvalue, principal eigenvector) pair
"""
vector_size = operator.size # input dimension of operator
if init_vec is None:
vec = torch.rand(vector_size)
else:
vec = init_vec
vec = utils.maybe_fp16(vec, fp16)
if use_gpu:
vec = vec.cuda()
prev_lambda = 0.0
prev_vec = utils.maybe_fp16(torch.randn_like(vec), fp16)
for i in range(steps):
prev_vec = vec / (torch.norm(vec) + 1e-6)
new_vec = utils.maybe_fp16(operator.apply(vec),
fp16) - momentum * prev_vec
# need to handle case where we end up in the nullspace of the operator.
# in this case, we are done.
if torch.norm(new_vec).item() == 0.0:
return 0.0, new_vec
lambda_estimate = vec.dot(new_vec).item()
diff = lambda_estimate - prev_lambda
vec = new_vec.detach() / torch.norm(new_vec)
if lambda_estimate == 0.0: # for low-rank
error = 1.0
else:
error = np.abs(diff / lambda_estimate)
utils.progress_bar(i, steps, "power iter error: %.4f" % error)
if error < error_threshold:
break
prev_lambda = lambda_estimate
return lambda_estimate, vec
def get_full_hessian(loss, param):
# from https://discuss.pytorch.org/t/compute-the-hessian-matrix-of-a-network/15270/3
# modified from hessian_eigenthings repo. api follows hessian.hessian
hessian_size = param.numel()
hessian = torch.zeros(hessian_size, hessian_size)
loss_grad = torch.autograd.grad(loss,
param,
create_graph=True,
retain_graph=True,
only_inputs=True)[0].view(-1)
for idx in range(hessian_size):
clear_output(wait=True)
progress_bar(idx, hessian_size,
"full hessian columns: %d of %d" % (idx, hessian_size))
grad2rd = torch.autograd.grad(loss_grad[idx],
param,
create_graph=False,
retain_graph=True,
only_inputs=True)
hessian[idx] = grad2rd[0].view(-1)
return hessian.cpu().data.numpy()
plt.ylabel("eigenvalue (log)")
plt.legend()
plt.suptitle("Hessian Spectrum Full Space")
plt.tight_layout(pad=0.6)
plt.savefig(join(savedir, "Hessian_sep_cls%d.jpg"%class_id))
# plt.show()
print("Spent %.1f sec from start" % (time() - T00))
#%% Interpolation in the full space
img_all = None
for eigi in range(50): #eigvects.shape[1]
interp_codes = LExpMap(ref_vect.cpu().numpy(), eigvects[:, -eigi-1], 11, (-2.5, 2.5))
img_list = G.visualize(torch.from_numpy(interp_codes).float().cuda()).cpu()
img_all = img_list if img_all is None else torch.cat((img_all, img_list), dim=0)
clear_output(wait=True)
progress_bar(eigi, 50, "ploting row of page: %d of %d" % (eigi, 256))
imggrid = make_grid(img_all, nrow=11)
PILimg = ToPILImage()(imggrid)#.show()
PILimg.save(join(savedir, "eigvect_full_cls%d.jpg"%class_id))
#% Interpolation in the class space
img_all = None
for eigi in range(50): # eigvects_clas.shape[1]
interp_class = LExpMap(classvec.cpu().numpy(), eigvects_clas[:, -eigi-1], 11, (-2.5, 2.5))
interp_codes = np.hstack((noisevec.cpu().numpy().repeat(11, axis=0), interp_class, ))
img_list = G.visualize(torch.from_numpy(interp_codes).float().cuda()).cpu()
img_all = img_list if img_all is None else torch.cat((img_all, img_list), dim=0)
clear_output(wait=True)
progress_bar(eigi, 50, "ploting row of page: %d of %d" % (eigi, 128))
imggrid = make_grid(img_all, nrow=11)