batch scoring in cpp

Start by reproducing the two paths in the issue with the ResNet symbol and parameter files, then trace the C++ NDArray::Load, Symbol::SimpleBind, and executor setup calls used by each method. Compare the resulting argument maps and batch handling, and define done as identifying the segmentation-fault cause and documenting which batch-scoring approach is valid and why.

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();