nghiapq77 / face-recognition-cpp-tensorrt Goto Github PK

Hi,
After compute similarity, some scores are above 1. This part of code get person and score.

std::tuple<std::vector<std::string>, std::vector<float>> ArcFaceIR50::getOutputs(float *output_sims) {
    /*
        Get person corresponding to maximum similarity score based on cosine similarity matrix.
    */
    std::vector<std::string> names;
    std::vector<float> sims;
    for (int i = 0; i < croppedFaces.size(); ++i) {
        int argmax = std::distance(output_sims + i * classCount, std::max_element(output_sims + i * classCount, output_sims + (i + 1) * classCount));
        float sim = *(output_sims + i * classCount + argmax);
        std::string name = classNames[argmax];
        names.push_back(name);
        sims.push_back(sim);
    }
    return std::make_tuple(names, sims);
}

I put a cout, printing sim and name. name is the user ID in database and sim is a float, but get numbers liks 122.717. Is it possible to get these values ? Why ?
Another question is, How do you get the sim value ? May you explain me. Why compute memory position in that way?

Thanks,

This a excellent project, the "config.json" have some parameters：

"input_embeddingsFile": "../data/test.json",
"input_numImagesFile": "../data/test_count.txt"

Can you provide these files? Thanks

arcface-ir50 weights is to big,how repalce it with arcface mobilefacenet?some code can supply?

Hi

I wanted to ask if there's any reason that you don't align the detected faces before delivering them to the Feature Extractor?

[INFO] Init cuBLASLt cosine similarity calculator...
terminate called after throwing an instance of 'nlohmann::detail::type_error'
what(): [json.exception.type_error.305] cannot use operator[] with a numeric argument with object
Aborted (core dumped)
运行环境：Jetson Xavier NX，cuda10.2，cudnn8.0.0

Hello , thanks for share code , it is working good ... i have a question , how we could get landmarks ?