better tests

CMakeLists.txt

@@ -13,8 +13,8 @@ add_library(tkDNN SHARED src/Layer.cpp src/LayerWgs.cpp
     src/Network.cpp src/utils.cpp)
 target_link_libraries(tkDNN kernels ${CUDA_LIBRARIES} ${CUDA_CUBLAS_LIBRARIES} -lcudnn)
 
-add_executable(tkDNNtest tests/test.cpp)
-target_link_libraries(tkDNNtest tkDNN)
+add_executable(test_simple tests/test/test.cpp)
+target_link_libraries(test_simple tkDNN)
 
-add_executable(mnist tests/mnist/test.cpp)
-target_link_libraries(mnist tkDNN)
+add_executable(test_mnist tests/mnist/test.cpp)
+target_link_libraries(test_mnist tkDNN)

tests/build_models.sh

@@ -0,0 +1,11 @@
+#!/bin/bash
+echo "build test Model"
+cd test
+python test_model.py
+cd ..
+cd mnist
+python mnist_model.py
+cd ..
+echo "export weights"
+python weights_exporter.py test/net.h5 --output test/layers
+python caffe_weights_exporter.py mnist/lenet.prototxt mnist/lenet.caffemodel --output mnist/layers

tests/caffe_weights_exporter.py

@@ -14,7 +14,6 @@
         help='Path to caffemodel file')
 
     parser.add_argument('--output', type=str, help="output directory", default="layers")
-    parser.add_argument('--test_db', type=str, help="input db to test", default=None)
 
     args = parser.parse_args()
 

tests/mnist/mnist_model.py

@@ -0,0 +1,32 @@
+#!/usr/bin/env python
+# mail:    admin@9crk.com
+# author:  9crk.from China.ShenZhen
+# time:    2017-03-22
+
+import caffe
+import numpy as np
+import cv2
+import sys
+import Image
+import matplotlib.pyplot as plt
+
+model = 'lenet.prototxt';
+weights = 'lenet.caffemodel';
+net = caffe.Net(model,weights,caffe.TEST);
+caffe.set_mode_gpu()
+img = np.array(np.random.rand(28,28), dtype=np.float32)
+#revert the image,and normalize it to 0-1 range
+
+print "INPUT: ", img
+img.tofile("input.bin", format="f")
+print "SHAPE: ", np.shape(img)
+out = net.forward_all(data=np.asarray([img]))
+
+out = out[out.keys()[0]]
+print out
+print np.shape(out)
+out.tofile("output.bin", format="f")
+#print out['prob'][0]
+#print out['prob'][0].argmax()
+
+

tests/mnist/test.cpp

@@ -0,0 +1,58 @@
+#include<iostream>
+#include "tkdnn.h"
+
+const char *input_bin   = "../tests/mnist/input.bin";
+const char *c0_bin      = "../tests/mnist/layers/Convolution0.bin";
+const char *c0_bias_bin = "../tests/mnist/layers/Convolution0.bias.bin";
+const char *c1_bin      = "../tests/mnist/layers/Convolution1.bin";
+const char *c1_bias_bin = "../tests/mnist/layers/Convolution1.bias.bin";
+const char *d2_bin      = "../tests/mnist/layers/InnerProduct2.bin";
+const char *d2_bias_bin = "../tests/mnist/layers/InnerProduct2.bias.bin";
+const char *d3_bin      = "../tests/mnist/layers/InnerProduct3.bin";
+const char *d3_bias_bin = "../tests/mnist/layers/InnerProduct3.bias.bin";
+const char *output_bin   = "../tests/mnist/output.bin";
+
+int main() {
+
+    // Network layout
+    tkDNN::Network net;
+    tkDNN::dataDim_t dim(1, 1, 28, 28, 1);
+    tkDNN::Layer *l;
+    l = new tkDNN::Conv2d     (&net, dim, 20, 5, 5, 1, 1, c0_bin, c0_bias_bin);
+    l = new tkDNN::Pooling    (&net, l->output_dim, 2, 2, 2, 2, tkDNN::POOLING_MAX);
+    l = new tkDNN::Conv2d     (&net, l->output_dim, 50, 5, 5, 1, 1, c1_bin, c1_bias_bin);
+    l = new tkDNN::Pooling    (&net, l->output_dim, 2, 2, 2, 2, tkDNN::POOLING_MAX);
+    l = new tkDNN::Dense      (&net, l->output_dim, 500, d2_bin, d2_bias_bin);
+    l = new tkDNN::Activation (&net, l->output_dim, CUDNN_ACTIVATION_RELU);
+    l = new tkDNN::Dense      (&net, l->output_dim, 10, d3_bin, d3_bias_bin);
+    l = new tkDNN::Softmax    (&net, l->output_dim);
+
+    // Load input
+    value_type *data;
+    value_type *input_h;
+    readBinaryFile(input_bin, dim.tot(), &input_h, &data);
+
+    printDeviceVector(dim.tot(), data);
+    dim.print(); //print initial dimension
+
+    TIMER_START
+
+    // Inference
+    data = net.infer(dim, data);
+
+    TIMER_STOP
+    dim.print();   
+
+    // Print result
+    std::cout<<"\n======= RESULT =======\n";
+    printDeviceVector(dim.tot(), data);
+
+    // Print real test
+    std::cout<<"\n==== CHECK RESULT ====\n";
+    value_type *out;
+    value_type *out_h;
+    readBinaryFile(output_bin, dim.tot(), &out_h, &out);
+    printDeviceVector(dim.tot(), out);
+
+    return 0;
+}

tests/test/test.cpp

@@ -0,0 +1,54 @@
+#include<iostream>
+#include "tkdnn.h"
+
+const char *input_bin   = "../tests/test/input.bin";
+const char *c0_bin      = "../tests/test/layers/conv0.bin";
+const char *c0_bias_bin = "../tests/test/layers/conv0.bias.bin";
+const char *c1_bin      = "../tests/test/layers/conv1.bin";
+const char *c1_bias_bin = "../tests/test/layers/conv1.bias.bin";
+const char *d2_bin      = "../tests/test/layers/dense2.bin";
+const char *d2_bias_bin = "../tests/test/layers/dense2.bias.bin";
+const char *output_bin   = "../tests/test/output.bin";
+
+int main() {
+
+    // Network layout
+    tkDNN::Network net;
+    tkDNN::dataDim_t dim(1, 1, 10, 10, 1);
+    tkDNN::Layer *l;
+    l = new tkDNN::Conv2d     (&net, dim, 2, 4, 4, 2, 2, c0_bin, c0_bias_bin);
+    l = new tkDNN::Activation (&net, l->output_dim, CUDNN_ACTIVATION_RELU);
+    l = new tkDNN::Conv2d     (&net, l->output_dim, 4, 2, 2, 1, 1, c1_bin, c1_bias_bin);
+    l = new tkDNN::Activation (&net, l->output_dim, CUDNN_ACTIVATION_RELU);
+    l = new tkDNN::Flatten      (&net, l->output_dim);
+    l = new tkDNN::Dense      (&net, l->output_dim, 4, d2_bin, d2_bias_bin);
+    l = new tkDNN::Activation (&net, l->output_dim, CUDNN_ACTIVATION_RELU);
+
+    // Load input
+    value_type *data;
+    value_type *input_h;
+    readBinaryFile(input_bin, dim.tot(), &input_h, &data);
+
+    printDeviceVector(dim.tot(), data);
+    dim.print(); //print initial dimension
+
+    TIMER_START
+
+    // Inference
+    data = net.infer(dim, data); dim.print();
+
+
+    TIMER_STOP
+
+    // Print result
+    std::cout<<"\n======= RESULT =======\n";
+    printDeviceVector(dim.tot(), data);
+
+    // Print real test
+    std::cout<<"\n==== CHECK RESULT ====\n";
+    value_type *out;
+    value_type *out_h;
+    readBinaryFile(output_bin, dim.tot(), &out_h, &out);
+    printDeviceVector(dim.tot(), out);
+    return 0;
+}

tests/test_model.py → tests/test/test_model.py

@@ -7,7 +7,6 @@
 from keras.models import Sequential, Model
 from keras.layers import Cropping2D
 import keras.backend.tensorflow_backend as KTF
-from weights_exporter import *
 
 def dense_model():
     model = Sequential()
@@ -17,6 +16,8 @@ def dense_model():
                            bias_initializer='random_uniform', activation="relu"))
     model.add(Convolution2D(4, (2, 2), subsample=(1, 1), 
                            bias_initializer='random_uniform', activation="relu"))
+    model.add(Flatten())
+    model.add(Dense(4, bias_initializer='random_uniform', activation="relu"))    
     sgd = keras.optimizers.Adam(lr=1e-4, decay=1e-8)
     model.compile(optimizer=sgd, loss="mse")
     return model
@@ -26,9 +27,7 @@ def dense_model():
     print "DATA FORMAT: ", keras.backend.image_data_format()
 
     model = dense_model()
-    wg = model.get_weights()
-    export_conv2d("conv0", wg[0], wg[1])
-    export_conv2d("conv1", wg[2], wg[3])
+    model.save("net.h5")
 
     grid = np.random.rand(10,10)
     X = grid[None,:,:]
@@ -41,3 +40,4 @@ def dense_model():
     print np.shape(r)  
     print "Result: ", r
     print "Result shape: ", np.shape(r) 
+    r.tofile("output.bin", format="f")

tests/weights_exporter.py

@@ -99,9 +99,7 @@ def get_session(gpu_fraction=0.5):
     parser = argparse.ArgumentParser(description='KERAS WEIGHTS EXPORTER TO CUDNN')
     parser.add_argument('model',type=str,
         help='Path to model h5 file. Model should be on the same path.')
-    parser.add_argument('layers', type=str, help="layers list [ dense, conv2d ]", nargs='+')
     parser.add_argument('--output', type=str, help="output directory", default="layers")
-    parser.add_argument('--test_db', type=str, help="input db to test", default=None)
 
     args = parser.parse_args()
 
@@ -119,34 +117,17 @@ def get_session(gpu_fraction=0.5):
 
     num = 0
     name_num = 0
-    for i in args.layers:
-        if i == "conv3d":
+    for l in model.layers:
+        name = l.name        
+        if name.startswith("conv3d"):
             export_conv3d(args.output + "/conv" + str(name_num), weights[num], weights[num+1])
-        elif i == "conv2d":
+        elif name.startswith("conv2d"):
             export_conv2d(args.output + "/conv" + str(name_num), weights[num], weights[num+1])
-        elif i == "dense":
+        elif name.startswith("dense"):
             export_dense(args.output + "/dense" + str(name_num), weights[num], weights[num+1])
         else:
-            print "error: ", i, "is not a layer type"
-            break
+            print "skip:", name, "has no weights"
+            continue
         name_num += 1
         num += 2
 
-    if args.test_db != None:
-        print "Test on db: ", args.test_db
-        db = lmdb.open(args.test_db, subdir=False, readonly=True, lock=False) 
-        txn = db.begin()
-
-        s = random.randint(0, txn.stat()["entries"]-1)
-        print "camp number: ", s
-        s = txn.get(str(s))
-        c = msgpack.unpackb(s)
-
-        print "Steer, throttle: ", c["actuators"]
-        print "Speed (m/s): ", c["speed"]
-        grid = np.asarray(c["bitmap"], np.float32)
-        i = np.array(grid.flatten(), dtype=np.float32)
-        i.tofile(args.output + "input.bin", format="f")
-        X = grid[None, :, :]
-
-        print "Prediction: ", model.predict(X)