mean.binaryproto problem why my mean.binaryproto 1KB without showing 3 channals values?

[WZhang]

i used compute_image_mean mean.binariproto

DEFINE_string(backend, "lmdb",

        "The backend {leveldb, lmdb} containing the images");

int main(int argc, char** argv) {

  ::google::InitGoogleLogging(argv[0]);

#ifdef USE_OPENCV

#ifndef GFLAGS_GFLAGS_H_

  namespace gflags = google;

#endif

  gflags::SetUsageMessage("Compute the mean_image of a set of images given by"

        " a leveldb/lmdb\n"

        "Usage:\n"

        "    compute_image_mean [FLAGS] INPUT_DB [OUTPUT_FILE]\n");

  /*gflags::ParseCommandLineFlags(&argc, &argv, true);*/

  argc = 3;

  argv[1] = "D:\\Deep Learning\\caffe_ssd_cuda8.0_vs2013\\caffe\\train_test_lmdb";

  argv[2] = "D:\\Deep Learning\\caffe_ssd_cuda8.0_vs2013\\caffe\\train_test_mean.binaryproto";

  //argv[1] = "D:\\Deep Learning\\caffe_ssd_cuda8.0_vs2013\\caffe\\00_datas\\car_1306\\trainval_lmdb\\";

  //argv[2] = "D:\\Deep Learning\\caffe_ssd_cuda8.0_vs2013\\caffe\\00_datas\\car_1306\\car1306_mean.binaryproto";

  if (argc < 2 || argc > 3) {

    gflags::ShowUsageWithFlagsRestrict(argv[0], "tools/compute_image_mean");

    return 1;

  }

  scoped_ptr<db::DB> db(db::GetDB(FLAGS_backend));

  db->Open(argv[1], db::READ);

  scoped_ptr<db::Cursor> cursor(db->NewCursor());

  BlobProto sum_blob;

  int count = 0;

  // load first datum

  Datum datum;

  datum.ParseFromString(cursor->value());

  if (DecodeDatumNative(&datum)) {

    LOG(INFO) << "Decoding Datum";

  }

  sum_blob.set_num(1);

  sum_blob.set_channels(datum.channels());

  sum_blob.set_height(datum.height());

  sum_blob.set_width(datum.width());

  const int data_size = datum.channels() * datum.height() * datum.width();

  int size_in_datum = std::max<int>(datum.data().size(),

                                    datum.float_data_size());

  for (int i = 0; i < size_in_datum; ++i) {

    sum_blob.add_data(0.);

  }

  LOG(INFO) << "Starting Iteration";

  while (cursor->valid()) {

    Datum datum;

    datum.ParseFromString(cursor->value());

    DecodeDatumNative(&datum);

    const std::string& data = datum.data();

    size_in_datum = std::max<int>(datum.data().size(),

        datum.float_data_size());

    CHECK_EQ(size_in_datum, data_size) << "Incorrect data field size " <<

        size_in_datum;

    if (data.size() != 0) {

      CHECK_EQ(data.size(), size_in_datum);

      for (int i = 0; i < size_in_datum; ++i) {

        sum_blob.set_data(i, sum_blob.data(i) + (uint8_t)data[i]);

      }

    } else {

      CHECK_EQ(datum.float_data_size(), size_in_datum);

      for (int i = 0; i < size_in_datum; ++i) {

        sum_blob.set_data(i, sum_blob.data(i) +

            static_cast<float>(datum.float_data(i)));

      }

    }

    ++count;

    if (count % 10000 == 0) {

      LOG(INFO) << "Processed " << count << " files.";

    }

    cursor->Next();

  }

  if (count % 10000 != 0) {

    LOG(INFO) << "Processed " << count << " files.";

  }

  for (int i = 0; i < sum_blob.data_size(); ++i) {

    sum_blob.set_data(i, sum_blob.data(i) / count);

  }

  // Write to disk

  if (argc == 3) {

    LOG(INFO) << "Write to " << argv[2];

    WriteProtoToBinaryFile(sum_blob, argv[2]);

  }

  const int channels = sum_blob.channels();

  const int dim = sum_blob.height() * sum_blob.width();

  std::vector<float> mean_values(channels, 0.0);

  LOG(INFO) << "Number of channels: " << channels;

  for (int c = 0; c < channels; ++c) {

    for (int i = 0; i < dim; ++i) {

      mean_values[c] += sum_blob.data(dim * c + i);

    }

    LOG(INFO) << "mean_value channel [" << c << "]:" << mean_values[c] / dim;

  }

#else

  LOG(FATAL) << "This tool requires OpenCV; compile with USE_OPENCV.";

#endif  // USE_OPENCV

  return 0;

}

get only train_test_mean.binaryproto without 3 channal values(3 channal image)

used convert_annoset to get train_test_lmdb

DEFINE_bool(gray, false,

    "When this option is on, treat images as grayscale ones");

DEFINE_bool(shuffle, false,

    "Randomly shuffle the order of images and their labels");

DEFINE_string(backend, "lmdb",

    "The backend {lmdb, leveldb} for storing the result");

DEFINE_string(anno_type, "detection",

    "The type of annotation {classification, detection}.");

DEFINE_string(label_type, "xml",

    "The type of annotation file format.");

DEFINE_string(label_map_file, "D:/Deep Learning/caffe_ssd_cuda8.0_vs2013/caffe/VOC0712/labelmap_voc.prototxt","A file with LabelMap protobuf message.");

//"D:\\Deep Learning\\caffe_ssd_cuda8.0_vs2013\\caffe\\00_datas\\car_1306\\car.prototxt"

DEFINE_bool(check_label, true,

    "When this option is on, check that there is no duplicated name/label.");

DEFINE_int32(min_dim, 0,

    "Minimum dimension images are resized to (keep same aspect ratio)");

DEFINE_int32(max_dim, 0,

    "Maximum dimension images are resized to (keep same aspect ratio)");

DEFINE_int32(resize_width, 512, "Width images are resized to");

DEFINE_int32(resize_height, 512, "Height images are resized to");

DEFINE_bool(check_size, true,

    "When this option is on, check that all the datum have the same size");

DEFINE_bool(encoded, true,

    "When this option is on, the encoded image will be save in datum");

DEFINE_string(encode_type, "jpg",

    "Optional: What type should we encode the image as ('png','jpg',...).");

int main(int argc, char** argv) {

#ifdef USE_OPENCV

  ::google::InitGoogleLogging(argv[0]);

  // Print output to stderr (while still logging)

  FLAGS_alsologtostderr = 1;

#ifndef GFLAGS_GFLAGS_H_

  namespace gflags = google;

#endif

  argc = 4;

  argv[1] = "D:/Deep Learning/caffe_ssd_cuda8.0_vs2013/caffe/";

  argv[2] = "D:/Deep Learning/caffe_ssd_cuda8.0_vs2013/caffe/train_test.txt";

  argv[3] = "D:/Deep Learning/caffe_ssd_cuda8.0_vs2013/caffe/train_test_lmdb";

train_test.txt

VOC2007/JPEGImages/000257.jpg VOC2007/Annotations/000257.xml

D:/Deep Learning/caffe_ssd_cuda8.0_vs2013/caffe/

D:/Deep Learning/caffe_ssd_cuda8.0_vs2013/caffe/train_test.txt

D:/Deep Learning/caffe_ssd_cuda8.0_vs2013/caffe/train_test_lmdb

I0418 15:37:15.999755  2340 convert_annoset.cpp:143] A total of 1 images.

I0418 15:37:16.016299  2340 db_lmdb.cpp:40] Opened lmdb D:/Deep Learning/caffe_ssd_cuda8.0_vs2013/caffe/train_test_lmdb

I0418 15:37:16.062636  2340 convert_annoset.cpp:222] Processed 1 files.