Can you take a look at the LibTorch model?

21fa417e3fb06e56040c · May 18, 2022, 8:13am

Hi everyone , I made a toy project. The project is a node JS web server using Libtorch.
I used a pre-trained model and the classification results are always constant. What could be the problem?

github.com

kerry-Cho/nodeJS-addon-example/blob/14b9a5702f6ad5d51eed6d0ed04a87303227e12d/src/classifier/classifier.cpp#L69

      
        
                    torch::Tensor out = torch::softmax(model->forward(img_tensor), 1);
            
            
        std::tuple<torch::Tensor, torch::Tensor> result = torch::max(out, 1);
            
            
        torch::Tensor prob = std::get<0>(result);
                    torch::Tensor index = std::get<1>(result);
            
            
        auto probability = prob.accessor<float,1>();
                    auto idx = index.accessor<int64_t, 1>();       
                    
                    return Napi::String::New(env, string_format("{ \"Class\": \"%d\", \"Probability\": \"%.3f\" }",
                            idx[0], probability[0]));
            
            
    } catch( std::exception& ex) {
                    return Napi::String::New(env, ex.what());
                }
            }