def test_sub_anttena(row, colomn, length):
center_x = -57.75 + (row * 2 + length - 1) / 2.0 * 16.5
center_y = -9.5 - (colomn * 2 + length - 1) / 2.0 * 16.5
subAntenna = SquareArray(length, center_y, center_x)
print(subAntenna.antenna_cor)
subAntennaIndex = numpy.zeros(shape=length * length)
for i in range(length * length):
subAntennaIndex[i] = (i // length + colomn) * 8 + i % length + row
offset = numpy.loadtxt(r'offsetfromstatic.txt')
offset = torch.Tensor(offset).view(1, -1)
trainPath = r'D:\dataset\data2'
trainDataset = FileDataSet.FileDataset(trainPath + r'\traindata.txt',
trainPath + r'\trainlabel.txt')
inputs, labels = trainDataset[:]
inputs = inputs[:, subAntennaIndex]
offset = offset[:, subAntennaIndex]
inputs = inputs - offset
labels = labels / 10
inputs = inputs.view(-1, 9, 1)
w = 360
h = 90
Al = torch.linspace(0, w - 1, w).view(1, w) / 180.0 * numpy.pi
Al = torch.matmul(torch.ones(h, 1), Al).view(1, 1, w * h)
Be = torch.linspace(0, h - 1, h).view(h, 1) / 180.0 * numpy.pi
Be = torch.matmul(
Be,
torch.ones(1, w),
).view(1, 1, w * h)
altruth = numpy.arctan2(labels[:, 1].numpy() - center_y,
labels[:, 0].numpy() - center_x) / numpy.pi * 180
betruth = numpy.arcsin((labels[:, 2].numpy() + 2) / numpy.sqrt(
(labels[:, 1].numpy() - center_y) * (labels[:, 1].numpy() - center_y) +
(labels[:, 0].numpy() - center_x) * (labels[:, 0].numpy() - center_x) +
(labels[:, 2].numpy() + 2) *
(labels[:, 2].numpy() + 2))) / numpy.pi * 180
index = 10
print(betruth[index])
print(altruth[index])
#plt.imshow(p0[index,:].view(h,w))
for i in range(labels.shape[0]):
index = i
a = subAntenna.p0(Al, Be, inputs[index, :, :].view(1, 9,
1)).view(h,
w).numpy()
print(
str(betruth[index]) + " " + str(altruth[index]) + " " +
str(numpy.unravel_index(a.argmax(), a.shape)[0]) + " " +
str(numpy.unravel_index(a.argmax(), a.shape)[1]))
def cal_offset(refanttena, trainPath):
center_x = -57.54 + (0 * 2 + 8 - 1) / 2.0 * 16.44
center_y = 57.54 - (0 * 2 + 8 - 1) / 2.0 * 16.44
test = SquareArray(8, center_y, center_x)
trainDataset = FileDataSet.FileDataset(trainPath + r'\traindata.txt',
trainPath + r'\trainlabel2.txt')
inputs, labels = trainDataset[:]
# plt.plot(numpy.mod(inputs.numpy()[:, 24] - inputs.numpy()[:, 8], numpy.pi * 2))
# plt.show()
d = test.theta_theory_XYZ(
torch.Tensor(labels[:, 0] * 100).view(-1, 1),
torch.Tensor(labels[:, 1] * 100).view(-1, 1),
torch.Tensor(labels[:, 2] * 100).view(-1, 1))
print(d.shape)
print(inputs.shape)
# plt.plot(inputs.numpy()[:, 9])
# plt.figure()
#
# plt.plot(labels.cpu().numpy()[0::1, 0])
# plt.plot(labels.cpu().numpy()[0::1, 1])
# plt.plot(
# numpy.mod(inputs.numpy()[:, 1] - inputs.numpy()[:, 8], numpy.pi * 2))
# plt.plot(
# numpy.mod(inputs.numpy()[:, 0] - inputs.numpy()[:, 8], numpy.pi * 2))
# plt.plot(numpy.mod((d.numpy()[:, 0] - d.numpy()[:, 8]), numpy.pi * 2))
# plt.show()
offset = numpy.zeros(shape=(64, 1))
for i in range(64):
offset[i, 0] = numpy.median(
numpy.mod((inputs.numpy()[:, i] - inputs.numpy()[:, refanttena] -
(d.numpy()[:, i] - d.numpy()[:, refanttena])),
numpy.pi * 2))
# plt.plot(offset)
# plt.show()
numpy.savetxt('offsetfromstatic.txt', offset)
def cal_offset():
center_x = -57.75 + (0 * 2 + 8 - 1) / 2.0 * 16.5
center_y = -9.5 - (0 * 2 + 8 - 1) / 2.0 * 16.5
test = SquareArray(8, center_y, center_x)
trainPath = r'D:\dataset\datastatic'
trainDataset = FileDataSet.FileDataset(trainPath + r'\traindata.txt',
trainPath + r'\trainlabel.txt')
inputs, labels = trainDataset[:]
d = test.theta_theory_XYZ(
torch.Tensor(labels[:, 0] / 10).view(-1, 1),
torch.Tensor(labels[:, 1] / 10).view(-1, 1),
torch.Tensor(labels[:, 2] / 10).view(-1, 1))
print(d.shape)
print(inputs.shape)
offset = numpy.zeros(shape=(64, 1))
for i in range(64):
offset[i, 0] = numpy.median(
numpy.mod(
(inputs.numpy()[:, i] - (d.numpy()[:, i] - d.numpy()[:, 0])),
numpy.pi * 2))
plt.plot(offset)
plt.show()
numpy.savetxt('offsetfromstatic.txt', offset)
class MyLoss(nn.Module):
def __init__(self):
super(MyLoss, self).__init__()
self.cro = nn.CrossEntropyLoss()
def forward(self, pred, truth):
return self.cro(pred[:, :pm.OutputShape[0]],
truth[:, 0, 0].cuda()) + self.cro(
pred[:, pm.OutputShape[0]:], truth[:, 0, 1].cuda())
workMode = pm.learnMode
dataMode = pm.dataMode
testPath = r'D:\Documents\OptiTrack\7-2-1'
fileDataset = FileDataSet.FileDataset(testPath + r'\traindata.txt',
testPath + r'\trainlabel2.txt')
#fileDataset.Uniform()
#fileDataset.make_more(2,0.01)
trainloader = torch.utils.data.DataLoader(fileDataset,
batch_size=10,
shuffle=True,
num_workers=0)
model = CHAModule.MyNet1(64, pm.picWidth)
#modelAE = torch.load('c.core')
#core = CHAModule.MyNet3()
#criterion = nn.MSELoss()
criterion = nn.CrossEntropyLoss(size_average=True)
w = 64
h = 48
import torch
from src.core import CenterCamera, Parameters as pm, AntennaArray as aa
import torchvision
import math
from src.dataprocess import FileDataSet
TestMoade = pm.LearningMode.Regression
if __name__ == '__main__':
#import TrainTestbed
model = torch.load('a.core')
model.eval()
testPath = r'E:\DataTest'
testDataset = FileDataSet.FileDataset(testPath + r'\testdata.txt',
testPath + r'\testlabel.txt')
testloader = torch.utils.data.DataLoader(testDataset,
batch_size=1,
shuffle=False,
num_workers=0)
inputs, labels = testDataset[:]
square_array = aa.SquareArray()
cam = CenterCamera.Camera()
w = int(cam.s[0, 0])
h = int(cam.s[1, 0])
pixalX = torch.linspace(0, w - 1, w).view(1, w)
pixalX = torch.matmul(torch.ones(h, 1), pixalX).view(1, w * h)
pixalY = torch.linspace(0, h - 1, h).view(h, 1)
pixalY = torch.matmul(
pixalY,
torch.ones(1, w),
).view(1, w * h)
import torch
from torch.autograd import Variable
import matplotlib.pyplot as plt
from src.core import Parameters as pm
from src.dataprocess import FileDataSet
TestMoade = pm.LearningMode.Regression
if __name__ == '__main__':
#import TrainTestbed
model = torch.load('b.core')
model.eval()
testPath = r'E:\Data7'
testDataset = FileDataSet.FileDatasetRNN(testPath + r'\traindata.txt',
testPath + r'\trainlabel.txt', 10)
testloader = torch.utils.data.DataLoader(testDataset,
batch_size=1,
shuffle=False,
num_workers=0)
for i, data in enumerate(testloader, 0):
if i % 100 != 0:
continue
inputs, labels = data
if torch.cuda.is_available():
inputs = inputs.cuda()
labels = labels.cuda()
outputs = model(Variable(inputs))
a = outputs.view(-1, 3072).cpu().detach().numpy()
b = labels.cpu().numpy()
print(b)
#plt.plot(b / 10)
def test_sub_anttena_Pixal(row, colomn, length, trainPath):
center_x = -57.54 + (colomn * 2 + length - 1) / 2.0 * 16.44
center_y = 57.54 - (row * 2 + length - 1) / 2.0 * 16.44
subAntenna = SquareArray(length, center_y, center_x)
print(subAntenna.antenna_cor)
subAntennaIndex = numpy.zeros(shape=length * length)
for i in range(length * length):
subAntennaIndex[i] = (i // length + colomn) * 8 + i % length + row
offset = numpy.loadtxt(r'offsetfromstatic.txt')
offset = torch.Tensor(offset).view(1, -1)
trainDataset = FileDataSet.FileDataset(trainPath + r'\traindata.txt',
trainPath + r'\trainlabel2.txt')
inputs, labels = trainDataset[:]
inputs = inputs[:, subAntennaIndex]
offset = offset[:, subAntennaIndex]
inputs = inputs - offset
labels = labels * 100
inputs = inputs.view(-1, length * length, 1)
w = 600
h = 600
Al = torch.linspace(0, w - 1, w).view(1, w)
Al = torch.matmul(torch.ones(h, 1), Al).view(1, 1, w * h)
Be = torch.linspace(0, h - 1, h).view(h, 1)
Be = torch.matmul(
Be,
torch.ones(1, w),
).view(1, 1, w * h)
camera = CenterCamera.Camera
Al, Be = camera.getAlBefromPixal(camera, Al, Be)
altruth = numpy.arctan2(labels[:, 1].numpy() - center_y,
labels[:, 0].numpy() - center_x) / numpy.pi * 180
betruth = numpy.arcsin((labels[:, 2].numpy() + 2) / numpy.sqrt(
(labels[:, 1].numpy() - center_y) * (labels[:, 1].numpy() - center_y) +
(labels[:, 0].numpy() - center_x) * (labels[:, 0].numpy() - center_x) +
(labels[:, 2].numpy() + 2) *
(labels[:, 2].numpy() + 2))) / numpy.pi * 180
index = 10
print(betruth[index])
print(altruth[index])
xtruth, ytruth = camera.getPixalFromAlBe(camera, altruth, betruth)
b = list()
#plt.imshow(p0[index,:].view(h,w))
imagepath = trainPath + r'\AOAPix\aoamap' + str(row) + '-' + str(
colomn) + '-' + str(length)
folder = os.path.exists(imagepath)
if not folder: # 判断是否存在文件夹如果不存在则创建为文件夹
os.makedirs(imagepath)
for i in range(0, labels.shape[0], 10):
index = i
a = subAntenna.p0(Al, Be, inputs[i, :, :].view(1, length * length,
1)).view(h, w)
torchvision.utils.save_image(
a, imagepath + '\\' + str(int(i / 10)) + '.jpg')
a = a.numpy()
b.append([
xtruth[index], ytruth[index],
numpy.unravel_index(a.argmax(), a.shape)[1],
numpy.unravel_index(a.argmax(), a.shape)[0]
])
#plt.show()
b = numpy.array(b)
numpy.savetxt(imagepath + r'\b.txt', numpy.array(b))
numpy.savetxt(imagepath + r'\label.txt', labels[::10, ].numpy())
import torch
from torch.autograd import Variable
import numpy as np
import matplotlib.pyplot as plt
from src.core import Parameters as pm
from src.dataprocess import FileDataSet
TestMoade = pm.LearningMode.Regression
if __name__ == '__main__':
#import TrainTestbed
# model = torch.load('c.core')
#model.eval()
trainPath = r'D:\dataset\data3'
trainDataset = FileDataSet.FileDataset(trainPath + r'\traindata.txt',
trainPath + r'\trainlabel.txt')
trainloader = torch.utils.data.DataLoader(trainDataset,
batch_size=1,
shuffle=False,
num_workers=0)
testPath = r'D:\dataset\data1'
testDataset = FileDataSet.FileDataset(testPath + r'\traindata.txt',
testPath + r'\trainlabel.txt')
testloader = torch.utils.data.DataLoader(testDataset,
batch_size=1,
shuffle=False,
num_workers=0)
inputs, labels = testDataset[:]
inputs2, labels2 = trainDataset[:]
#inputs = model(Variable(inputs.cuda())).detach().cpu()
def TestOneTag(epc):
model = torch.load('a.core')
model.eval()
testPath = r'E:\Data20'
testDataset = FileDataSet.FileDataset(
testPath + r'\traindata' + str(epc) + '.txt',
testPath + r'\trainlabel' + str(epc) + '.txt')
testloader = torch.utils.data.DataLoader(testDataset,
batch_size=1,
shuffle=False,
num_workers=0)
criterion = nn.MSELoss()
modelAE = torch.load('c.core')
# randindex = torch.linspace(1,100,100)#np.random.randint(0, 80, size=[10])
# for i in randindex:
# inputs, labels = testDataset[int(i)]
# labels=labels.view(1,1,2)
# one_hot = torch.zeros(1, 640).scatter_(1, labels.data[:,:,0],1)
# inputs, labels = Variable(inputs), Variable(one_hot.view(-1, 640))
# if torch.cuda.is_available():
# inputs = inputs.cuda()
# labels = labels.cuda()
# outputs = core(Variable(inputs ))
# scal = (torch.Tensor([640, 480]).view(1, 2)).cuda()
#
# #print(outputs.data * scal)
# print(outputs.data)
# print(labels)
# print('========')
# inputs, labels = testDataset[:]
# outputs=core(Variable(inputs.cuda()))
# #scal = (torch.Tensor([640, 480]).view(1, 2)).cuda()
# #np.savetxt('a.txt', (outputs.data*scal).cpu().numpy(), fmt='%.6f')
# np.savetxt('a.txt', (outputs.data).cpu().numpy(), fmt='%.6f')
# for epoch in range(1): # loop over the dataset multiple times
#
# running_loss = 0.0
# for i, data in enumerate(testloader, 0):
# # get the inputs
# inputs, labels = data
# one_hot = torch.zeros(labels.size(0), 640).scatter_(1, labels.data[:, :, 0], 1)
# inputs, labels = Variable(inputs), Variable(one_hot.view(-1, 1, 640))
#
# # wrap them in Variable
# #inputs, labels = Variable(inputs ), Variable(labels /torch.Tensor([640,480]).view(1,2))
# if torch.cuda.is_available():
# inputs = inputs.cuda()
# labels = labels.cuda()
#
#
# outputs = core(inputs)
# loss = criterion(outputs, labels)
#
#
# # print statistics
# running_loss += loss.data
# if i % 20 == 19: # print every 2000 mini-batches
# print('[%d, %5d] loss: %.5f' %
# (epoch + 1, i + 1, running_loss / 20))
# running_loss = 0.0
# # break
#
# print('Finished Training')
if TestMoade == pm.LearningMode.Classification1LabelHeatMap:
inputs, labels = testDataset[:]
if torch.cuda.is_available():
inputs = inputs.cuda()
labels = labels.cuda()
# outputs = core(Variable(modelAE.encoder(inputs)))
outputs = model(Variable((inputs)))
a = outputs.cpu().detach().numpy().transpose()
# x = F.softmax(torch.Tensor(a[:pm.OutputShape[0],:]),0)
# values, indices = torch.max(x, 0)
# y = F.softmax(torch.Tensor(a[pm.OutputShape[0]:, :]), 0)
# values2, indices2 = torch.max(y, 0)
# index =torch.cat((indices.view(-1,1),indices2.view(-1,1)),1)
# np.savetxt('a.txt',index.numpy().astype(int))
# np.savetxt('b.txt', x.numpy())
b = labels.cpu().numpy()[0::10, 0]
plt.plot(b / 10)
# plt.figure()
# for i in range(100):
# plt.imshow(a[:, i * 10].reshape(48, 64))
# plt.show()
# plt.imshow(a[:, 1].reshape(48, 64))
plt.imshow(np.sum(a[:, 0::10].reshape(48, 64, -1), 0))
temp = np.unravel_index(np.argmax(a[:, 0::10], 0), (48, 64))
plt.plot(temp[1])
plt.figure()
b = labels.cpu().numpy()[0::10, 1]
plt.plot(b / 10)
# plt.figure()
# for i in range(100):
# plt.imshow(a[:, i * 10].reshape(48, 64))
# plt.show()
# plt.imshow(a[:, 1].reshape(48, 64))
plt.imshow(np.sum(a[:, 0::10].reshape(48, 64, -1), 1))
temp = np.unravel_index(np.argmax(a[:, 0::10], 0), (48, 64))
plt.plot(temp[0])
plt.figure()
r1 = labels.cpu().numpy()[0::10, 1] / 10 - temp[0]
r2 = labels.cpu().numpy()[0::10, 0] / 10 - temp[1]
plt.hist(np.abs(r1), 100)
plt.figure()
plt.hist(np.abs(r2), 100)
plt.figure()
plt.hist(np.abs(np.sqrt(r1 * r1 + r2 * r2)), 100)
plt.figure()
plt.plot(np.sqrt(r1 * r1 + r2 * r2))
print(np.mean(np.abs(r1)))
print(np.mean(np.abs(r2)))
temp = np.unravel_index(np.argmax(a[:, :], 0), (48, 64))
temp2 = np.hstack(
(temp[1].reshape(-1, 1) * 10, temp[0].reshape(-1, 1) * 10))
np.savetxt(testPath + r'\modelResult' + str(epc) + '.txt', temp2)
# for i, data in enumerate(testloader, 0):
# if i % 100 != 0:
# continue
# inputs, labels = data
# if torch.cuda.is_available():
# inputs = inputs.cuda()
# labels = labels.cuda()
# outputs = core(Variable(inputs))
# a = outputs.view(-1, 3072).cpu().detach().numpy()
# b = labels.cpu().numpy()
# print(b)
# plt.imshow(a[:, :].reshape(48, 64))
# plt.show()
elif TestMoade == pm.LearningMode.Regression:
inputs, labels = testDataset[:]
if torch.cuda.is_available():
inputs = inputs.cuda()
labels = labels.cuda()
outputs = model(Variable((inputs)))
b = labels.cpu().numpy()[:, 0]
plt.figure()
plt.plot(b)
plt.plot(outputs.detach().cpu().numpy()[:, 0] * 640)
plt.figure()
b = labels.cpu().numpy()[:, 1]
plt.plot(b)
plt.plot((outputs.detach().cpu().numpy()[:, 1]) * 240 + 240)
temp2 = np.hstack(
(outputs.detach().cpu().numpy()[:, 0].reshape(-1, 1) * 640,
outputs.detach().cpu().numpy()[:, 1].reshape(-1, 1) * 240 +
240))
np.savetxt(testPath + r'\modelResult' + str(epc) + '.txt', temp2)
elif TestMoade == pm.LearningMode.Classification2LabelsOneHot:
inputs, labels = testDataset[:]
if torch.cuda.is_available():
inputs = inputs.cuda()
labels = labels.cuda()
outputs = model(Variable(inputs))
a = outputs.cpu().detach().numpy().transpose()
x = F.softmax(torch.Tensor(a[pm.OutputShape[0]:, :]), 0)
# values, indices = torch.max(x, 0)
# y = F.softmax(torch.Tensor(a[pm.OutputShape[0]:, :]), 0)
plt.imshow(x * x)
b = labels.cpu().numpy()[:, 1]
plt.plot(b)
aa = np.argmax(x[:, 0::1], 0)
plt.plot(aa.numpy())
plt.show()