TensorRT 2 is not able to load the models trained with DIGITS 6.0.0

Accelerated Computing GPU-Accelerated Libraries
1

Hi everyone,
The problem is when I try to load a model that is trained with DIGITS 6.0.0 and NVcaffe 0.16.4, I encounter this error:

conv1: ERROR - 32-bit weights not found for 32-bit model
conv1: ERROR - 32-bit weights not found for 32-bit model
Parameter check failed at: Network.cpp::addConvolution::32, condition: kernelWeights.values != NULL
error parsing layer type Convolution index 1

But older models are fine (I don’t know the exact version of DIGITS with which they were trained but it’s 5.IDon’tKnowWhat and the version NVcaffe that was used was probably 0.15.14.

This is the code I use to build the models:

nvinfer1::IBuilder* builder = nvinfer1::createInferBuilder(gLogger);

// Creating the network definition structure
nvinfer1::INetworkDefinition* network = builder->createNetwork();
nvcaffeparser1::ICaffeParser* parser = nvcaffeparser1::createCaffeParser();
const nvcaffeparser1::IBlobNameToTensor* blobNameToTensor = parser->parse(DeployFile, CaffeModel,
                                                                          *network, nvinfer1::DataType::kFLOAT);

// Marking the output blob
network->markOutput(*blobNameToTensor->find(CoverageOutputBlob));
network->markOutput(*blobNameToTensor->find(BboxOutputBlob));

// Building the engine from the network definition:
builder->setMaxBatchSize(BatchSize);
builder->setMaxWorkspaceSize(1 << 20);
nvinfer1::ICudaEngine* engine = builder->buildCudaEngine(*network);

// Serializing the engine to a memory block and saving it for later use
nvinfer1::IHostMemory* SerializedEngine = engine->serialize();