1)Is it possible get each the layer's top labels (e.g: ip1,ip2,conv1,conv2) in c++?
If my layer is
layer {
name: "relu1_1"
type: "Input"
top: "pool1"
input_param {
shape: {
dim:1
dim: 1
dim: 28
dim: 28
}
}
}
I want to get the top label which in my case is "pool1"
I searched the examples provided, but I couldn't find anything. currently I'm able to get only the layers names and layer type by the following commands,
cout << "Layer name:" << "'" << net_->layer_names()[layer_index]<<endl;
cout << "Layer type: " << net_->layers()[layer_index]->type()<<endl;
2) Where can I find the tutorials or the examples which explains most used API's for using caffe framework using c++?
Thankyou in advance.
Look at Net class in doxygen:
const vector< vector< Blob< Dtype > * > > all_ tops = net_->top_vecs(); // get "top" of all layers
Blob<Dtype>* ptop = all_tops[layer_index][0]; // pointer to top blob of layer
If you want the layer's name, you can
const string layer_name = net_->layer_names()[layer_index];
You can access all sorts of names/data using net_ interface, just read the doc!
Related
How can I parse the following YAML file using yaml-cpp?
scene:
- camera:
film:
width: 800
height: 600
filename: "out.svg"
- shape:
name: "muh"
I tried:
#include <yaml-cpp/yaml.h>
int main() {
YAML::Node root_node = YAML::LoadFile("Scenes/StanfordBunny.flatland.yaml");
// throws an exception
int value = root_node["scene"]["camera"]["film"]["width"].as<int>();
}
How can I get the value of the width attribute?
How can I get the name of the shape attribute?
The "-" in front of camera means it is an array of objects. So my guess would be:
root_node["scene"][0]["camera"]["film"]["width"].as<int>();
I'm trying to deploy a caffe model containing a RNN layer. The issue I'm having is how to compute the output from the network. My assumption was that I could call
net->Forward();
to update the network and then
net->output_blobs()[0]->mutable_cpu_data()[x];
once every timestep to read the output. However, using a constant input and then running "net-Forward()" multiple times does not affect the output as one would expect. I've tried to use different weights/biases, which changes the output, but no matter what configuration I'm using the output will still be static. Does anyone know what the proper procedure for deploying caffe RNNs with C++ is?
Edit:
This was tested with a single neuron RNN layer like below.
model.prototxt:
layer {
name: "input"
type: "Input"
top: "states"
input_param {
shape: {
dim: 1
dim: 1
}
}
}
input: "clip"
input_shape { dim: 1 dim: 1 dim: 1}
layer {
name: "rnn"
type: "RNN"
top: "rnn"
bottom: "clip"
bottom: "states"
recurrent_param {
num_output: 1
}
}
And the .cpp:
caffe::Blob<float>* input_layer = test_net->input_blobs()[0];
float* input_data;
input_data = input_layer->mutable_cpu_data();
input_data[0] = 1.0;
for (int i=0; i<5; i++)
{
test_net->Forward();
cout << "Ouput: " << net->output_blobs().back()->mutable_cpu_data()[0] << endl;
}
I have been trying to output a yaml file using YAML::Emitter. For instance, I need something like this to be my yaml file.
annotations:
- run:
type: range based
attributes:
start_frame:
frame_number: 25
end_frame:
frame_number: 39
So far, using my code
for (auto element : obj)
{
basenode = YAML::LoadFile(filePath); //loading a file throws exception when it is not a valid yaml file
//Check if metadata is already there
if (!basenode["file_reference"])
{
writeMetaData(element.getGttStream(), element.getTotalFrames(), element.getFileHash());
}
annotationNode["annotations"].push_back(element.getGestureName());
annotationNode["type"] = "range based";
output << annotationNode;
attributesNode["attributes"]["start_frame"]["frame_number"] = element.getStartFrame();
attributesNode["attributes"]["end_frame"]["frame_number"] = element.getEndFrame();
output << typeNode;
output << attributesNode;
ofs.open(filePath, std::ios_base::app);
ofs << std::endl << output.c_str();
}
I am getting an output like this
annotations:
- run
type: range based
---
attributes:
start_frame:
frame_number: 26
end_frame:
frame_number: 57
I want the "type" and "attributes" under the recently pushed sequence item into the "annotations" and subsequently the same for all the following nodes.
I even tried using something like this
annotationNode[0][type] = "range based"
and the output was like this
0: type: "range based"
How do i get the recently pushed item in the sequence "annotations"?
If you're building up your root node, annotationNode, then just build it up and output it once. You wouldn't need to write either the typeNode or attributesNode to the emitter. For example, you might write
YAML::Node annotationNode;
for (auto element : obj) {
YAML::Node annotations;
annotations["name"] = element.getGestureName();
annotations["type"] = ...;
annotations["attributes"] = ...;
annotationNode["annotations"] = annotations;
}
output << annotationNode;
I'm writing C++ code using CAFFE to predict a single (for now) image. The image has already been preprocessed and is in .png format. I have created a Net object and read in the trained model. Now, I need to use the .png image as an input layer and call net.Forward() - but can someone help me figure out how to set the input layer?
I found a few examples on the web, but none of them work, and almost all of them use deprecated functionality. According to: Berkeley's Net API, using "ForwardPrefilled" is deprecated, and using "Forward(vector, float*)" is deprecated. API indicates that one should "set input blobs, then use Forward() instead". That makes sense, but the "set input blobs" part is not expanded on, and I can't find a good C++ example on how to do that.
I'm not sure if using a caffe::Datum is the right way to go or not, but I've been playing with this:
float lossVal = 0.0;
caffe::Datum datum;
caffe::ReadImageToDatum("myImg.png", 1, imgDims[0], imgDims[1], &datum);
caffe::Blob< float > *imgBlob = new caffe::Blob< float >(1, datum.channels(), datum.height(), datum.width());
//How to get the image data into the blob, and the blob into the net as input layer???
const vector< caffe::Blob< float >* > &result = caffeNet.Forward(&lossVal);
Again, I'd like to follow the API's direction of setting the input blobs and then using the (non-deprecated) caffeNet.Forward(&lossVal) to get the result as opposed to making use of the deprecated stuff.
EDIT:
Based on an answer below, I updated to include this:
caffe::MemoryDataLayer<unsigned char> *memory_data_layer = (caffe::MemoryDataLayer<unsigned char> *)caffeNet.layer_by_name("input").get();
vector< caffe::Datum > datumVec;
datumVec.push_back(datum);
memory_data_layer->AddDatumVector(datumVec);
but now the call to AddDatumVector is seg faulting.. I wonder if this is related to my prototxt format? here's the top of my prototxt:
name: "deploy"
input: "data"
input_shape {
dim: 1
dim: 3
dim: 100
dim: 100
}
layer {
name: "conv1"
type: "Convolution"
bottom: "data"
top: "conv1"
I base this part of the question on this discussion about a "source" field being important in the prototxt...
caffe::Datum datum;
caffe::ReadImageToDatum("myImg.png", 1, imgDims[0], imgDims[1], &datum);
MemoryDataLayer<float> *memory_data_layer = (MemoryDataLayer<float> *)caffeNet->layer_by_name("data").get();
memory_data_layer->AddDatumVector(datum);
const vector< caffe::Blob< float >* > &result = caffeNet.Forward(&lossVal);
Something like this could be useful. Here you will have to use MemoryData layer as the input layer. I am expecting the layer name to be named data.
The way of using datum variable may not be correct. If my memory is correct, I guess, you have to use a vector of datum data.
I think this should get you started.
Happy brewing. :D
Here is an excerpt from my code located here where I used Caffe in my C++ code. I hope this helps.
Net<float> caffe_test_net("models/sudoku/deploy.prototxt", caffe::TEST);
caffe_test_net.CopyTrainedLayersFrom("models/sudoku/sudoku_iter_10000.caffemodel");
// Get datum
Datum datum;
if (!ReadImageToDatum("examples/sudoku/cell.jpg", 1, 28, 28, false, &datum)) {
LOG(ERROR) << "Error during file reading";
}
// Get the blob
Blob<float>* blob = new Blob<float>(1, datum.channels(), datum.height(), datum.width());
// Get the blobproto
BlobProto blob_proto;
blob_proto.set_num(1);
blob_proto.set_channels(datum.channels());
blob_proto.set_height(datum.height());
blob_proto.set_width(datum.width());
int size_in_datum = std::max<int>(datum.data().size(),
datum.float_data_size());
for (int ii = 0; ii < size_in_datum; ++ii) {
blob_proto.add_data(0.);
}
const string& data = datum.data();
if (data.size() != 0) {
for (int ii = 0; ii < size_in_datum; ++ii) {
blob_proto.set_data(ii, blob_proto.data(ii) + (uint8_t)data[ii]);
}
}
// Set data into blob
blob->FromProto(blob_proto);
// Fill the vector
vector<Blob<float>*> bottom;
bottom.push_back(blob);
float type = 0.0;
const vector<Blob<float>*>& result = caffe_test_net.Forward(bottom, &type);
What about:
Caffe::set_mode(Caffe::CPU);
caffe_net.reset(new caffe::Net<float>("your_arch.prototxt", caffe::TEST));
caffe_net->CopyTrainedLayersFrom("your_model.caffemodel");
Blob<float> *your_blob = caffe_net->input_blobs()[0];
your_blob->set_cpu_data(your_image_data_as_pointer_to_float);
caffe_net->Forward();
I'm using the mongocxx driver and I am considering keeping the query results given in BSON as a data holder in a couple of objects instead of parsing the BSON to retrieve the values and then discard it.
This would make some sense "if" I can edit the BSON on the fly. I couldn't find anything in the bsoncxx driver documentation besides the builder that would allow me to manipulate a bsoncxx document/value/view/element after it's been constructed.
As an example, imagine that I have something like this
fruit["orange"];
where fruit is a bsoncxx::document::element
I can get the value by using one of the .get_xxx operators.
What I can't find is something like
fruit["orange"] = "ripe";
Is there a way of doing this, or the idea behind the builder is "just" to create a query to give to the database?
There was a question with same theme, see here
So, bsoncxx objects seem to be immutable, and we have to re-create them if we need to edit them.. :(
I've written a really bad solution which re-creates document from scratch
But this is a solution, I guess.
std::string bsoncxx_string_viewToString(core::v1::string_view gotStringView) {
std::stringstream convertingStream;
convertingStream << gotStringView;
return std::move(convertingStream.str());
}
std::string b_utf8ToString(bsoncxx::types::b_utf8 gotB_utf8) {
return std::move(bsoncxx_string_viewToString(core::v1::string_view(gotB_utf8)));
}
template <typename T>
bsoncxx::document::value editBsoncxx(bsoncxx::document::view documentToEdit, std::string keyToEdit, T newValue, bool appendValueIfKeyNotExist = true) {
auto doc = bsoncxx::builder::stream::document{};
std::string currentKey;
for (auto i : documentToEdit) {
currentKey = bsoncxx_string_viewToString(i.key());
if (currentKey == keyToEdit) {
doc << keyToEdit << newValue;
appendValueIfKeyNotExist = false;
} else {
doc << currentKey << i.get_value();
}
}
if (appendValueIfKeyNotExist) // Maybe this would be better with documentToEdit.find(key), but I don't know how to check if iterator is past-the-end
//If there is a way to check if bsoncxx contains key, we can achieve ~o(log(n)) [depending on 'find key' implementation] which is better than o(n)
doc << keyToEdit << newValue;
return doc.extract();
}
Usage:
auto doc = document{} << "foo0" << "bar0" << "foo1" << 1 << "foo2" << 314 << finalize;
std::cout << bsoncxx::to_json(doc) << std::endl << std::endl;
doc = editBsoncxx<std::string> (doc.view(), "foo1", "edited"); //replace "foo1" with string "edited"
doc = editBsoncxx<int>(doc.view(), "baz_noappend", 123, false); //do nothing if key "baz_noappend" is not found. <- if key-existance algorythm will be applied, we'd spend about o(lob(n)) here, not o(n)
doc = editBsoncxx<int>(doc.view(), "baz_append", 123, true); //key will not be found => it'll be appended which is default behaviour
std::cout << bsoncxx::to_json(doc) << std::endl;
Result:
{
"foo0" : "bar0",
"foo1" : 1,
"foo2" : 314
}
{
"foo0" : "bar0",
"foo1" : "edited",
"foo2" : 314,
"baz_append" : 123
}
So, in your case you can use
fruit = editBsoncxx<std::string>(fruit.view(), "orange", "ripe");
But, again, see already-mentioned related question you're right when saying that
the idea behind the builder is "just" to create a query to give to the database?
I think, the solution will be "do not edit documents".
also you can write something like type-converter from bsoncxx to other json storing fomat (for example, rapidjson)
Beware of {value:"valid_json"}: bsoncxx::to_json does not add backslashes to quote signs in values => injection can be made.