mmr/tkDNN
Archived
1
0
Fork 0
Code Issues 46 Pull Requests 12 Releases 4 Activity

works but it need cleaning

Browse Source
This commit is contained in:
Francesco Gatti
2020-02-16 16:28:39 +01:00
parent 8736a8c3da
commit 0ac292ea48
9 changed files with 219 additions and 71 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tests/imuodom/imuodom.cpp
+15 -11
View File
@@ -6,7 +6,6 @@ const char *i1_bin = "../tests/imuodom/layers/input1.bin";
const char *i2_bin = "../tests/imuodom/layers/input2.bin";
const char *o0_bin = "../tests/imuodom/layers/output0.bin";
const char *o1_bin = "../tests/imuodom/layers/output1.bin";
const char *output_bin = "../tests/imuodom/layers/output.bin";
const char *c0_bin = "../tests/imuodom/layers/conv1d_7.bin";
const char *c1_bin = "../tests/imuodom/layers/conv1d_8.bin";
@@ -17,6 +16,7 @@ const char *c5_bin = "../tests/imuodom/layers/conv1d_12.bin";
const char *l0_bin = "../tests/imuodom/layers/bidirectional_3.bin";
const char *l1_bin = "../tests/imuodom/layers/bidirectional_4.bin";
const char *d0_bin = "../tests/imuodom/layers/dense_3.bin";
const char *d1_bin = "../tests/imuodom/layers/dense_4.bin";
int main() {
@@ -51,10 +51,14 @@ int main() {
tk::dnn::Layer *concat_l[3] = { &x0_2, &x1_2, &x2_2 };
tk::dnn::Route concat (&net, concat_l, 3);
//tk::dnn::LSTM lstm0(&net, 128, true, l0_bin);
//tk::dnn::LSTM lstm1(&net, 128, false, l1_bin);
//tk::dnn::Dense d0 (&net, 3, d0_bin);
tk::dnn::LSTM lstm0(&net, 128, true, l0_bin);
tk::dnn::LSTM lstm1(&net, 128, false, l1_bin);
tk::dnn::Dense d0 (&net, 3, d0_bin);
tk::dnn::Layer *lstm1_l[1] = { &lstm1 };
tk::dnn::Route lstm1_link (&net, lstm1_l, 1);
tk::dnn::Dense d1 (&net, 4, d1_bin);
net.print();
dnnType *data;
@@ -66,12 +70,12 @@ int main() {
TIMER_STOP
// Print real test
//std::cout<<"\n==== CHECK RESULT ====\n";
//dnnType *out;
//dnnType *out_h;
//readBinaryFile(output_bin, dim.tot(), &out_h, &out);
//checkResult(dim.tot(), data, out);
printDeviceVector(100, data);
std::cout<<"\n==== CHECK RESULT ====\n";
dnnType *out0, *out1;
dnnType *out0_h, *out1_h;
readBinaryFile(o0_bin, d0.output_dim.tot(), &out0_h, &out0);
readBinaryFile(o1_bin, d1.output_dim.tot(), &out1_h, &out1);
checkResult(d0.output_dim.tot(), d0.dstData, out0);
checkResult(d1.output_dim.tot(), d1.dstData, out1);
return 0;
}
tests/imuodom/infer.py
+14 -11
View File
@@ -34,30 +34,33 @@ if __name__ == '__main__':
[yhat_delta_p, yhat_delta_q] = model.predict([x_angle, x_gyro, x_acc], batch_size=1, verbose=1)
layer_name = 'bidirectional_3'
intermediate_layer_model = Model(inputs=model.input,
outputs=model.get_layer(layer_name).output)
intermediate_output = intermediate_layer_model.predict([x_angle, x_gyro, x_acc])
#layer_name = 'dense_4'
#intermediate_layer_model = Model(inputs=model.input,
# outputs=model.get_layer(layer_name).output)
#intermediate_output = intermediate_layer_model.predict([x_angle, x_gyro, x_acc])
x_angle = np.array([x_angle])
x_gyro = np.array([x_gyro])
x_acc = np.array([x_acc])
intermediate_output = np.array([intermediate_output])
#intermediate_output = np.array([intermediate_output])
x_angle = x_angle.transpose(0, 3, 1, 2)
x_gyro = x_gyro.transpose(0, 3, 1, 2)
x_acc = x_acc.transpose(0, 3, 1, 2)
intermediate_output = intermediate_output.transpose(0, 3, 1, 2)
#intermediate_output = intermediate_output.transpose(0, 3, 1, 2)
#print("Aggregate:")
#print(intermediate_output.tolist())
print("x0: ", np.shape(x_angle))
print("out: ",np.shape(intermediate_output))
#print("out: ",np.shape(intermediate_output))
x_angle = np.array(x_angle.flatten(), dtype=np.float32)
x_gyro = np.array(x_gyro.flatten(), dtype=np.float32)
x_acc = np.array(x_acc.flatten(), dtype=np.float32)
yhat_delta_p = np.array(yhat_delta_p.flatten(), dtype=np.float32)
yhat_delta_q = np.array(yhat_delta_q.flatten(), dtype=np.float32)
intermediate_output = np.array(intermediate_output.flatten(), dtype=np.float32)
#intermediate_output = np.array(intermediate_output.flatten(), dtype=np.float32)
f = open("layers/input0.bin", mode='wb')
@@ -70,6 +73,6 @@ if __name__ == '__main__':
bin_write(f, yhat_delta_p)
f = open("layers/output1.bin", mode='wb')
bin_write(f, yhat_delta_q)
f = open("layers/output.bin", mode='wb')
bin_write(f, intermediate_output)
#f = open("layers/output.bin", mode='wb')
#bin_write(f, intermediate_output)
tests/simple/test_model.py
+10 -5
View File
@@ -2,6 +2,7 @@ import keras
import numpy as np
from keras.models import Sequential
from keras.layers import Input, Dense, Activation, Flatten, Dropout, ELU, Reshape, Lambda, Conv1D
from keras.layers import Bidirectional, CuDNNLSTM
from keras.layers.convolutional import Convolution2D, Convolution3D
from keras.layers.pooling import MaxPooling2D, MaxPooling3D, AveragePooling3D
from keras.models import Sequential, Model
@@ -17,9 +18,11 @@ def bin_write(f, data):
f.write(bin)
def create_model():
x1 = Input((6, 16), name='x1')
x1 = Input((3, 8), name='x1')
conv = Conv1D(4, 2)(x1)
model = Model([x1], [conv])
lstm = Bidirectional(CuDNNLSTM(5, return_sequences=True))(conv)
lstm2 = Bidirectional(CuDNNLSTM(5, return_sequences=False))(lstm)
model = Model([x1], [lstm2])
model.summary()
return model
@@ -30,14 +33,16 @@ if __name__ == '__main__':
model = create_model()
model.save("net.hdf5")
x = np.random.rand(1,1,6,16)
np.random.seed(2)
x = np.random.rand(1,1,3,8)
r = model.predict( x[0], batch_size=1)
r = np.array([r])
r = np.array([r])
x = x.transpose(0, 3, 1, 2)
r = r.transpose(0, 3, 1, 2)
#r = r.transpose(0, 3, 1, 2)
print("in: ", np.shape(x))
print("out: ", np.shape(r))
print("output: ", r.tolist())
x = np.array(x.flatten(), dtype=np.float32)
f = open("input.bin", mode='wb')
tests/simple/test_simple.cpp
+7 -1
View File
@@ -3,14 +3,20 @@
const char *input_bin = "../tests/simple/input.bin";
const char *c0_bin = "../tests/simple/layers/conv1d_1.bin";
const char *l1_bin = "../tests/simple/layers/bidirectional_1.bin";
const char *l2_bin = "../tests/simple/layers/bidirectional_2.bin";
const char *output_bin = "../tests/simple/output.bin";
int main() {
// Network layout
tk::dnn::dataDim_t dim(1, 16, 1, 6);
tk::dnn::dataDim_t dim(1, 8, 1, 3);
tk::dnn::Network net(dim);
tk::dnn::Conv2d l0(&net, 4, 1, 2, 1, 1, 0, 0, c0_bin);
tk::dnn::LSTM l1(&net, 5, true, l1_bin);
tk::dnn::LSTM l2(&net, 5, false, l2_bin);
net.print();
// Load input
dnnType *data;
tests/weights_exporter.py
+37 -7
View File
@@ -42,17 +42,47 @@ def export_layer(name, weights, bias):
bin_write(f, bias)
print ("WEIGHTS saved\n")
def export_bidir(name, params):
def export_bidir(name, params, paramsb):
print ("######## EXPORT", name, "LAYER ########")
f = open(name + ".bin", mode='wb')
print("FORWARD")
ker = params[0]
rec_ker = params[1]
bias = params[2]
print ("export kernels: ", np.shape(ker))
units = np.shape(ker)[1] // 4
bin_write(f, ker[:,:units])
bin_write(f, ker[:,units:units*2])
bin_write(f, ker[:,units*2:units*3])
bin_write(f, ker[:,units*3:])
print ("export recurrent kernels: ", np.shape(rec_ker))
bin_write(f, rec_ker[:,:units])
bin_write(f, rec_ker[:,units:units*2])
bin_write(f, rec_ker[:,units*2:units*3])
bin_write(f, rec_ker[:,units*3:])
print ("export kernels: ", np.shape(ker))
bin_write(f, bias)
print("WEIGHTS saved\n")
for w in params:
#w = w.transpose()
print(np.shape(w))
bin_write(f, w)
print("BACKWARD")
ker = paramsb[0]
rec_ker = paramsb[1]
bias = paramsb[2]
print ("export kernels: ", np.shape(ker))
units = np.shape(ker)[1] // 4
bin_write(f, ker[:,:units])
bin_write(f, ker[:,units:units*2])
bin_write(f, ker[:,units*2:units*3])
bin_write(f, ker[:,units*3:])
print ("export recurrent kernels: ", np.shape(rec_ker))
bin_write(f, rec_ker[:,:units])
bin_write(f, rec_ker[:,units:units*2])
bin_write(f, rec_ker[:,units*2:units*3])
bin_write(f, rec_ker[:,units*3:])
print ("export kernels: ", np.shape(ker))
bin_write(f, bias)
print("WEIGHTS saved\n")
#https://github.com/fchollet/keras/wiki/Converting-convolution-kernels-from-Theano-to-TensorFlow-and-vice-versa
@@ -100,7 +130,7 @@ if __name__ == '__main__':
export_layer(args.output + "/" + name, wgs[0], wgs[1])
elif name.startswith("bidirectional"):
wgs = l.forward_layer.get_weights()
export_bidir(args.output + "/" + name, wgs)
export_bidir(args.output + "/" + name, l.forward_layer.get_weights(), l.backward_layer.get_weights())
else:
print ("skip:", name, "has no weights")
continue