In [1]:
import os
os.chdir('/home/renwh/gaze/renwh/caffe_with_cudnnv3/')

import sys
sys.path.insert(0,'./python')
import caffe

from pylab import *
%matplotlib inline


caffe.set_device(0)
caffe.set_mode_gpu()
solver = caffe.SGDSolver('examples/mymodel/02/lenet_solver.prototxt')

#You can choose to load your model status
#solver.restore('examples/mymodel/03/lenet_iter_1001.solverstate')
In [2]:
# each output is (batch size, feature dim, spatial dim)
[(k, v.data.shape) for k, v in solver.net.blobs.items()]
Out[2]:
[('data', (1000, 1, 36, 60)),
 ('label', (1000, 6)),
 ('gaze', (1000, 3)),
 ('headpose', (1000, 3)),
 ('conv1', (1000, 20, 32, 56)),
 ('pool1', (1000, 20, 16, 28)),
 ('conv2', (1000, 30, 12, 24)),
 ('pool2', (1000, 30, 6, 12)),
 ('conv3', (1000, 50, 2, 8)),
 ('pool3', (1000, 50, 1, 4)),
 ('flatdata', (1000, 200)),
 ('cat', (1000, 203)),
 ('ip3', (1000, 3)),
 ('loss', ())]
In [3]:
# just print the weight sizes (not biases)
[(k, v[0].data.shape) for k, v in solver.net.params.items()]
Out[3]:
[('conv1', (20, 1, 5, 5)),
 ('conv2', (30, 20, 5, 5)),
 ('conv3', (50, 30, 5, 5)),
 ('ip3', (3, 203))]
In [4]:
solver.net.forward()  # train net
solver.test_nets[0].forward()  # test net (there can be more than one)
Out[4]:
{'loss': array(0.5318098068237305, dtype=float32)}
In [5]:
# we use a little trick to tile the first eight images
imshow(solver.test_nets[0].blobs['data'].data[:8, 0].transpose(1, 0, 2).reshape(36, 8*60), cmap='gray')
print solver.net.blobs['label'].data[:8]

[[ -2.44513273e-01   5.20949736e-02  -9.68245506e-01  -5.07045567e-01
   -1.12138920e-01  -2.90884897e-02]
 [ -7.41908699e-02   2.27922529e-01  -9.70848620e-01  -1.28387764e-01
    1.65355857e-02   1.06296828e-03]
 [ -1.74087971e-01   3.04691344e-02  -9.84258592e-01  -9.52000245e-02
   -3.14195365e-01  -1.50917871e-02]
 [ -2.49744281e-02   1.77879885e-01  -9.83735263e-01  -7.38587156e-02
   -1.21144764e-02  -4.47588827e-04]
 [ -1.61419377e-01   5.79187945e-02  -9.85184848e-01  -1.06810793e-01
    1.42905980e-01   7.65229668e-03]
 [ -1.52415037e-01   2.09456533e-01  -9.65866268e-01  -5.29863574e-02
   -1.14266567e-01  -3.03129526e-03]
 [ -1.76816806e-02   6.62708879e-02  -9.97644961e-01  -6.35477304e-02
   -2.95568883e-01  -9.46362782e-03]
 [  1.79661021e-01   2.34958977e-01  -9.55257118e-01  -8.40480402e-02
    1.60711512e-01   6.77234307e-03]]
In [6]:
solver.step(1)
In [7]:
imshow(solver.net.params['conv1'][0].diff[:, 0].reshape(4, 5, 5, 5)
       .transpose(0, 2, 1, 3).reshape(4*5, 5*5), cmap='gray')
Out[7]:
<matplotlib.image.AxesImage at 0x7f16600943d0>

Show the conv1 weights pics.

Then, I will train the model, and log some information.

In [8]:
%%time
niter = 1000
test_interval = 25
# losses will also be stored in the log
train_loss = zeros(niter)
mean_error= zeros(int(np.ceil(niter / test_interval)))
output = zeros((niter, 8, 3))

# the main solver loop
for it in range(niter):
    solver.step(1)  # SGD by Caffe
    
    # store the train loss
    train_loss[it] = solver.net.blobs['loss'].data
    
    # store the output on the first test batch
    # (start the forward pass at conv1 to avoid loading new data)
    solver.test_nets[0].forward(start='conv1')
    output[it] = solver.test_nets[0].blobs['ip3'].data[:8]
    if it % test_interval == 0:
        # caculate the square error for each gaze vector
        solver.test_nets[0].forward()
        
        num_test = 100;
        sub_error = zeros((num_test, 3))
        square_error = zeros((num_test, 3))
        sum_square_error = zeros(num_test)
        for i in range(num_test):
            sub_error[i,:] = np.subtract(solver.test_nets[0].blobs['label'].data[i,:3]
                                         , solver.test_nets[0].blobs['ip3'].data[i])
            square_error = np.square(sub_error)
            sum_square_error = np.sum(square_error,1)
        mean_error[it // test_interval] = np.sum(sum_square_error,0)/num_test*180
        print 'Iteration', it, '. Mean error is', mean_error[it // test_interval]

Iteration 0 . Mean error is 121.252891377
Iteration 25 . Mean error is 5.44572881544
Iteration 50 . Mean error is 5.5682508211
Iteration 75 . Mean error is 5.1006434863
Iteration 100 . Mean error is 7.78466295475
Iteration 125 . Mean error is 4.26385671
Iteration 150 . Mean error is 6.50231969635
Iteration 175 . Mean error is 3.35166009118
Iteration 200 . Mean error is 3.18154300774
Iteration 225 . Mean error is 5.80954781502
Iteration 250 . Mean error is 3.13526394369
Iteration 275 . Mean error is 5.42981514199
Iteration 300 . Mean error is 2.84072761998
Iteration 325 . Mean error is 2.6534085312
Iteration 350 . Mean error is 4.06035709239
Iteration 375 . Mean error is 3.19245751223
Iteration 400 . Mean error is 5.77443953144
Iteration 425 . Mean error is 2.5737843768
Iteration 450 . Mean error is 2.73599987109
Iteration 475 . Mean error is 3.28580591441
Iteration 500 . Mean error is 2.82761177203
Iteration 525 . Mean error is 6.00738391138
Iteration 550 . Mean error is 2.50860575861
Iteration 575 . Mean error is 2.75102245027
Iteration 600 . Mean error is 2.90444281038
Iteration 625 . Mean error is 2.78774371646
Iteration 650 . Mean error is 5.59055560124
Iteration 675 . Mean error is 2.2716631448
Iteration 700 . Mean error is 3.10566709546
Iteration 725 . Mean error is 2.47598092939
Iteration 750 . Mean error is 2.55948991959
Iteration 775 . Mean error is 5.75543647197
Iteration 800 . Mean error is 2.36303983528
Iteration 825 . Mean error is 3.3922549448
Iteration 850 . Mean error is 2.64478828715
Iteration 875 . Mean error is 2.32434053005
Iteration 900 . Mean error is 5.27097275915
Iteration 925 . Mean error is 2.40158495176
Iteration 950 . Mean error is 3.54577662922
Iteration 975 . Mean error is 2.40263874427
CPU times: user 51.8 s, sys: 7.08 s, total: 58.9 s
Wall time: 58.9 s
In [9]:
_, ax1 = subplots()
ax2 = ax1.twinx()
ax1.plot(arange(niter), train_loss)
ax2.plot(test_interval * arange(len(mean_error)), mean_error, 'r')
ax1.set_xlabel('iteration')
ax1.set_ylabel('train loss')
ax2.set_ylabel('mean error')
Out[9]:
<matplotlib.text.Text at 0x7f165a048f90>

**show you the train loss curve.

In [10]:
num_test = 1000


# (start the forward pass at conv1 to avoid loading new data)
solver.test_nets[0].forward(start='conv1')
solver.test_nets[0].forward()

#figure(figsize=(10, 5))
#imshow(solver.test_nets[0].blobs['data'].data[:num_test, 0].transpose(1, 0, 2).reshape(36, num_test*60), cmap='gray')
    
# print the label and train result
#for i in range(num_test):
#    print solver.test_nets[0].blobs['label'].data[i,:3] ,'label<->ip3', solver.test_nets[0].blobs['ip3'].data[i]

print '--------------------------------------------------------------------------------------------------------------'
# caculate the square error for each gaze vector
sub_error = zeros((num_test, 3))
square_error = zeros((num_test, 3))
sum_square_error = zeros(num_test)
for i in range(num_test):
    sub_error[i,:] = np.subtract(solver.test_nets[0].blobs['label'].data[i,:3], solver.test_nets[0].blobs['ip3'].data[i])
    square_error = np.square(sub_error)
    sum_square_error = np.sum(square_error,1)
    #print sub_error[i,:],square_error[i,:],sum_square_error[i]
    #print sum_square_error[i],

print num_test,'test pic, mean error is ',np.sum(sum_square_error,0)/num_test*180,'degree'
_, ax1 = subplots()
ax1.plot(arange(num_test), sum_square_error*180,'bo', label='sampled')
ax1.set_xlabel('num_test')
ax1.set_ylabel('sum_square_error')
    

--------------------------------------------------------------------------------------------------------------
1000 test pic, mean error is  2.24705385909 degree
Out[10]:
<matplotlib.text.Text at 0x7f1659f93910>
In [11]:
imshow(solver.net.params['conv1'][0].diff[:, 0].reshape(4, 5, 5, 5)
       .transpose(0, 2, 1, 3).reshape(4*5, 5*5), cmap='gray')
Out[11]:
<matplotlib.image.AxesImage at 0x7f1659e91c10>
In [12]:
figure(figsize=(10, 5))
imshow(solver.net.params['conv2'][0].diff[:, 0].reshape(3, 10, 5, 5)
       .transpose(0, 2, 1, 3).reshape(3*5, 10*5), cmap='gray')
Out[12]:
<matplotlib.image.AxesImage at 0x7f1659e12f50>
In [13]:
figure(figsize=(10, 5))
imshow(solver.net.params['conv3'][0].diff[:, 0].reshape(5, 10, 5, 5)
       .transpose(0, 2, 1, 3).reshape(5*5, 10*5), cmap='gray')
Out[13]:
<matplotlib.image.AxesImage at 0x7f1659d4f390>
In [14]:
figure(figsize=(20, 10))
imshow(solver.test_nets[0].blobs['conv1'].data[:8, :].reshape(8,20,32,56)
           .transpose(0,2,1,3).reshape(8*32, 20*56), cmap='gray')
Out[14]:
<matplotlib.image.AxesImage at 0x7f1659c83290>
In [15]:
figure(figsize=(20, 10))
imshow(solver.test_nets[0].blobs['conv2'].data[:8, :].reshape(8, 30, 12, 24)
       .transpose(0,2,1,3).reshape(8*12, 30*24), cmap='gray')
Out[15]:
<matplotlib.image.AxesImage at 0x7f1659a6a890>
In [16]:
figure(figsize=(50, 25))
imshow(solver.test_nets[0].blobs['conv3'].data[:8, :].reshape(16, 25, 2, 8)
       .transpose(0,2,1,3).reshape(16*2, 25*8), cmap='gray')
Out[16]:
<matplotlib.image.AxesImage at 0x7f1659b9e590>
In [17]:
#solver.net.save('my_model.caffemodel') I do not know how to use this.
solver.snapshot() #SAVE MY MODEL IN THE DIR YOU DEFINE IN SOLVER FILE.
In [ ]: