Understanding a Prediction

[acpn]

Hy guys I'm doing a prediction with code of https://gist.github.com/onauparc/dd80907401b26b602885

this a code:

int main(int argc, char** argv)
{
// Set GPU
Caffe::set_mode(Caffe::GPU);
int device_id = 0;
Caffe::SetDevice(device_id);
LOG(INFO) << "Using GPU";

// Load net
Net net("/home/antonio/Models/Multipie/alexnet_test.prototxt", caffe::TEST);//<-- depends on caffe version

// Load pre-trained net
net.CopyTrainedLayersFrom("/home/antonio/Models/Multipie/caffenet_train_iter_120000.caffemodel");
// Load image
cv::Mat image = cv::imread("/home/antonio/multipie/test_complete/90_346_653212.png"); // or cat.jpg

// Set vector for image
vectorcv::Mat imageVector;
imageVector.push_back(image); // image is a cv::Mat, as I'm using #1416 //push_back()

// Set vector for label
vector labelVector;
labelVector.push_back(0);//push_back 0 for initialize purpose

// Net initialization
float loss = 0.0;

shared_ptr<MemoryDataLayer > memory_data_layer;

memory_data_layer = boost::static_pointer_cast<MemoryDataLayer >(net.layer_by_name("data_n"));

memory_data_layer->AddMatVector(imageVector,labelVector);

// Run ForwardPrefilled
vector<Blob*> results = net.ForwardPrefilled(&loss);

// Display result
const shared_ptr<Blob >& probs = net.blob_by_name("prob");
const float* probs_out = probs->cpu_data();

for (int i = 0 ; i < 3 ; i++)
{
LOG(INFO) << "Prob" << i << ": " << probs_out[i];
}

LOG(INFO) << "Max element is: " << *max_element(probs_out,probs_out+3);

// Display result2
const float* argmaxs = results[1]->cpu_data();

for (int i = 0; i < results[1]->num(); i++)
{
for (int j = 0; j < results[1]->height(); j++)
{
LOG(INFO) << "Image: "<< i << " class:" << argmaxs[i*results[1]->height() + j];
}
}
return 0;
}

and compile ok, but this is a resul when I run this code:

W0620 21:25:47.396652 30355 memory_data_layer.cpp:88] MemoryData does not transform array data on Reset()
I0620 21:25:47.403425 30355 test_model.cpp:66] Prob0: 6.28324e-31
I0620 21:25:47.403463 30355 test_model.cpp:66] Prob1: 7.08585e-28
I0620 21:25:47.403467 30355 test_model.cpp:66] Prob2: 1.03385e-23
I0620 21:25:47.403471 30355 test_model.cpp:69] Max element is: 1.03385e-23
I0620 21:25:47.403475 30355 test_model.cpp:78] Image: 0 class:0

This is strange because my accuracy stay between 92% and 97%, and image and class is alway 0

Any Idea?

thanks.

[jshfeng]

Thanks for sharing the problem. However, this tracker is for issues and bugs with Caffe. Please ask usage questions on the caffe-users list. https://groups.google.com/forum/#!forum/caffe-users