GstInference and GstEmbeddingMeta metadata

This metadata consists of a variable size elements array, that is filed with the embedding produced by the net, like facenet that uses a 128 elements array size.

GstEmbeddingMeta example

Fields

The facenet element and embedding overlay uses similar metadata as the classification plugins. GstEmbeddingMeta consist on the following fields:

field	type	description
num_dimensions	gint	The number of labels outputted by the model. This can vary from model to model. Facenet uses 128.
embedding	gdouble *	The embedding produced by the network

Access metadata

If you want to access this metadata from your custom Gstreamer element instead the process is fairly easy:

1. Add the Facenet element to your pipeline
2. Include GstInference metadata header: #include "gst/r2inference/gstinferencemeta.h"
3. Get a GstClassificationMeta object from the buffer: class_meta = (GstClassificationMeta *) gst_buffer_get_meta (frame->buffer, GST_CLASSIFICATION_META_API_TYPE);

FaceNet also raises a signal containing GstClassificationMeta, for details on how to use this signal please check the example applications section.

In the following section of code, we add the include like in point 2 and safe the GstClassificationMeta in the class_meta variable like in point 3.

#include "gst/r2inference/gstinferencemeta.h"
static void 
get_buffer(GstPadProbeInfo * info)
{
  GstBuffer *buffer;
  GstDetectionMeta *meta;
  buffer = gst_pad_probe_info_get_buffer (info);
  class_meta = (GstClassificationMeta *) gst_buffer_get_meta (buffer,
                  GST_DETECTION_META_API_TYPE);

  g_print ("Dimension: 0 has embedding %f\n", class_meta->embedding[0]);
}