batch scoring in cpp

Comience reproduciendo las dos rutas del issue con los archivos de símbolos y parámetros de ResNet, y luego rastree las llamadas de C++ NDArray::Load, Symbol::SimpleBind y de configuración del executor que utiliza cada método. Compare los mapas de argumentos resultantes y el manejo de batches, y defina como terminado identificar la causa del segmentation fault y documentar qué enfoque de batch-scoring es válido y por qué.

I am trying to load a pre-built model and do batch scoring, I tried the following two ways, the first one can produce some results, and the second one compiles successfully but runs with segmentation fault. Could you please let me know which way I should follow, and why the second way doesn't work? Thanks!
method 1:

/* Image size and channels */
int width = 224;
int height = 224;
int channels = 3;
int batch_size = 5;

Context ctx_dev(DeviceType::kCPU, 0);
map<string, NDArray> args_map;
map<string, NDArray> aux_map;
map<string, NDArray> parameters;

NDArray::Load("../Resnet/resnet-152-0000.params", 0, &parameters);

for (const auto &k : parameters) {
  if (k.first.substr(0, 4) == "aux:") {
    auto name = k.first.substr(4, k.first.size() - 4);
    aux_map[name] = k.second.Copy(ctx_dev);
  }
  if (k.first.substr(0, 4) == "arg:") {
    auto name = k.first.substr(4, k.first.size() - 4);
    args_map[name] = k.second.Copy(ctx_dev);
  }
}

auto net = Symbol::Load("../Resnet/resnet-152-symbol.json");

auto data_iter = MXDataIter("ImageRecordIter")
.SetParam("path_imglist","../caltech_256/caltech-256-60-train.lst")
.SetParam("path_imgrec","../caltech_256/caltech-256-60-train.rec")
.SetParam("data_shape", Shape(3, 224, 224))
.SetParam("batch_size", batch_size)
.SetParam("shuffle", 1)
.CreateDataIter();

while(data_iter.Next()){
auto batch = data_iter.GetData();
args_map["data"] = batch;
auto *exec = net.SimpleBind(ctx_dev, args_map);
exec->Forward(false);
auto outputs = exec->outputs[0].Copy(Context(kCPU, 0));
NDArray::WaitAll();
for (int i = 0; i <2; i++) {
cout << outputs.At(0, i) <<",";
}
cout << endl;
}
MXNotifyShutdown();

method 2:

int width = 224;
int height = 224;
int channels = 3;
int batch_size = 5;

Context ctx_dev(DeviceType::kCPU, 0);

map<string, NDArray> args_map;
map<string, NDArray> aux_map;

args_map["data"] = NDArray(Shape(batch_size, channels, width, height), ctx_dev);
args_map["label"] = NDArray(Shape(batch_size), ctx_dev);

auto net = Symbol::Load("../Resnet/resnet-152-symbol.json");

auto *exec = net.SimpleBind(ctx_dev, args_map);

NDArray::Load("../Resnet/resnet-152-0000.params", 0, &args_map);

auto data_iter = MXDataIter("ImageRecordIter")
  .SetParam("path_imglist","../caltech_256/caltech-256-60-train.lst")
  .SetParam("path_imgrec","../caltech_256/caltech-256-60-train.rec")
  .SetParam("data_shape", Shape(3, 224, 224))
  .SetParam("batch_size", batch_size)
  .SetParam("shuffle", 1)
  .CreateDataIter();

while(data_iter.Next()){
auto batch = data_iter.GetDataBatch();
batch.data.CopyTo(&args_map["data"]);
batch.label.CopyTo(&args_map["label"]);
exec->Forward(false);
NDArray::WaitAll();
}

delete exec;
MXNotifyShutdown();