def imagenet_norm(x):
mean = [0.485, 0.456, 0.406]
std = [0.299, 0.224, 0.225]
mean = torch.FloatTensor(mean).unsqueeze(0).unsqueeze(2).unsqueeze(3).to(
x.device)
std = torch.FloatTensor(std).unsqueeze(0).unsqueeze(2).unsqueeze(3).to(
x.device)
return (x - mean) / std
def get_preds_fromhm(hm):
max, idx = torch.max(
hm.view(hm.size(0), hm.size(1), hm.size(2) * hm.size(3)), 2)
idx += 1
preds = idx.view(idx.size(0), idx.size(1), 1).repeat(1, 1, 2).float()
preds[..., 0].apply_(lambda x: (x - 1) % hm.size(3) + 1)
preds[..., 1].add_(-1).div_(hm.size(2)).floor_().add_(1)
for i in range(preds.size(0)):
for j in range(preds.size(1)):
hm_ = hm[i, j, :]
pX, pY = int(preds[i, j, 0]) - 1, int(preds[i, j, 1]) - 1
if pX > 0 and pX < 63 and pY > 0 and pY < 63:
diff = torch.FloatTensor(
[hm_[pY, pX + 1] - hm_[pY, pX - 1],
hm_[pY + 1, pX] - hm_[pY - 1, pX]])
preds[i, j].add_(diff.sign_().mul_(.25))
preds.add_(-0.5)
return preds
import paddle import paddorch as paddorch import torch import numpy as np x = np.arange(2 * 768).reshape((1, 2, 768)) x_t = torch.repeat_interleave(torch.FloatTensor(x), repeats=4, dim=1) x_p = paddorch.repeat_interleave(paddorch.FloatTensor(x), repeats=4, dim=1) assert np.max(np.abs(x_t.numpy() - x_p.numpy())) < 0.001, "fail match torch" print(paddorch.repeat_interleave(x_p, 2)) y = paddorch.tensor([[1, 2], [3, 4]]) print(paddorch.repeat_interleave(y, 2)) print(paddorch.repeat_interleave(y, 3, dim=1))
import pickle
import paddorch
import numpy as np
a = paddorch.FloatTensor(np.array([ 0.9, 0.1]))
pickle.dump(a,open("tensor.pkl",'wb'))
b=pickle.load(open("tensor.pkl",'rb'))
print(b)
a = paddorch.LongTensor(np.array([ 9, 1]))
pickle.dump(a,open("tensor.pkl",'wb'))
b=pickle.load(open("tensor.pkl",'rb'))
print(b)
N_dim = 200
N_dim2 = N_dim
torch.manual_seed(0)
a = torch.randn(N_dim2, N_dim).to_sparse().requires_grad_(False)
a_dense = a.to_dense().numpy()
b = torch.randn(N_dim, N_dim2, requires_grad=True)
torch_y = torch.sparse.mm(a, b)
print(torch_y)
import paddorch
import paddle
import numpy as np
a = paddorch.sparse.FloatTensor(
paddorch.LongTensor(a._indices().detach().numpy()),
paddorch.FloatTensor(a._values().detach().numpy()), (N_dim2, N_dim))
b = paddorch.from_numpy(b.detach().numpy())
b_param = paddorch.nn.Parameter(b)
b.stop_gradient = False
a.values.stop_gradient = False
y = paddorch.sparse.mm(a, b)
print("max diff", np.max(np.abs(torch_y.detach().numpy() - y.numpy())))
c = a.to_dense()
import time
before = time.time()
for _ in range(6):
y = paddorch.sparse.mm(a, b)
# y = paddorch.mm(c, b )
y = torch.sparse.mm(a, b)
b=torch.cat([b,y],dim=1)
y.sum().backward()
after=time.time()
print("time:",after-before)
# print("max grad",torch.max(a.grad))
if device=="cuda":
import sys
sys.exit()
with fluid.dygraph.guard(place=place):
a=paddorch.sparse.FloatTensor(paddorch.LongTensor(np.stack([I,J]) ), paddorch.FloatTensor(V ) ,(N_dim2, N_dim))
b=paddorch.from_numpy(b_np )
b_param=paddorch.nn.Parameter(b)
b.stop_gradient=False
a.values.stop_gradient=False
import time
before=time.time()
for _ in range(6):
y = paddorch.sparse.mm(a, b )
b=paddorch.cat([b,y],dim=1)
break
y.sum().backward()
def np2tensor(image):
"""Converts numpy array to torch tensor."""
return torch.FloatTensor(image).permute(2, 0, 1) / 255 * 2 - 1
