Merge branch 'dev-0.6'

Author: GFallasRR

COPYING

@@ -5,7 +5,7 @@ dnl please read gstreamer/docs/random/autotools before changing this file
 dnl initialize autoconf
 dnl releases only do -Wall, git and prerelease does -Werror too
 dnl use a three digit version number for releases, and four for git/pre
-AC_INIT([GStreamer Inference],[0.5.0.1],[https://github.com/RidgeRun/gst-inference/issues],[gst-inference])
+AC_INIT([GStreamer Inference],[0.6.0.1],[https://github.com/RidgeRun/gst-inference/issues],[gst-inference])
 
 AG_GST_INIT
 

configure.ac

@@ -1,13 +1,16 @@
 plugin_LTLIBRARIES = libgstinference.la
 
 libgstinference_la_SOURCES =		\
+	gstinceptionv1.c 		\
 	gstinceptionv2.c		\
+	gstinceptionv3.c 		\
 	gstinceptionv4.c		\
 	gstinference.c			\
 	gsttinyyolov2.c			\
 	gsttinyyolov3.c                 \
 	gstfacenetv1.c                  \
-	gstresnet50v1.c
+	gstresnet50v1.c			\
+	gstmobilenetv2.c
 
 libgstinference_la_CFLAGS =		\
 	$(GST_CFLAGS)			\
@@ -37,9 +40,12 @@ libgstinference_la_LIBTOOLFLAGS =	\
 	$(GST_PLUGIN_LIBTOOLFLAGS)
 
 noinst_HEADERS =			\
+	gstinceptionv1.h 		\
 	gstinceptionv2.h		\
+	gstinceptionv3.h 		\
 	gstinceptionv4.h		\
 	gsttinyyolov2.h			\
 	gsttinyyolov3.h                 \
 	gstfacenetv1.h                  \
-	gstresnet50v1.h
+	gstresnet50v1.h			\
+	gstmobilenetv2.h

ext/r2inference/Makefile.am

@@ -42,10 +42,15 @@
 #include "gst/r2inference/gstinferencemeta.h"
 #include <string.h>
 #include <math.h>
+#include "gst/r2inference/gstinferencepreprocess.h"
+#include "gst/r2inference/gstinferencepostprocess.h"
+#include "gst/r2inference/gstinferencedebug.h"
 
 GST_DEBUG_CATEGORY_STATIC (gst_facenetv1_debug_category);
 #define GST_CAT_DEFAULT gst_facenetv1_debug_category
 
+#define MODEL_CHANNELS 3
+
 /* prototypes */
 static void gst_facenetv1_set_property (GObject * object,
     guint property_id, const GValue * value, GParamSpec * pspec);
@@ -195,66 +200,8 @@ static gboolean
 gst_facenetv1_preprocess (GstVideoInference * vi,
     GstVideoFrame * inframe, GstVideoFrame * outframe)
 {
-  gint i, j, pixel_stride, width, height, channels;
-  gfloat mean, std, variance, sum, normalized, R, G, B;
-
   GST_LOG_OBJECT (vi, "Preprocess");
-  channels = GST_VIDEO_FRAME_N_COMPONENTS (inframe);
-  pixel_stride = GST_VIDEO_FRAME_COMP_STRIDE (inframe, 0) / channels;
-  width = GST_VIDEO_FRAME_WIDTH (inframe);
-  height = GST_VIDEO_FRAME_HEIGHT (inframe);
-
-  sum = 0;
-  normalized = 0;
-
-  for (i = 0; i < height; ++i) {
-    for (j = 0; j < width; ++j) {
-      sum =
-          sum + (((guchar *) inframe->data[0])[(i * pixel_stride +
-                  j) * channels + 0]);
-      sum =
-          sum + (((guchar *) inframe->data[0])[(i * pixel_stride +
-                  j) * channels + 1]);
-      sum =
-          sum + (((guchar *) inframe->data[0])[(i * pixel_stride +
-                  j) * channels + 2]);
-    }
-  }
-
-  mean = sum / (float) (width * height * channels);
-
-  for (i = 0; i < height; ++i) {
-    for (j = 0; j < width; ++j) {
-      R = (gfloat) ((((guchar *) inframe->data[0])[(i * pixel_stride +
-                      j) * channels + 0])) - mean;
-      G = (gfloat) ((((guchar *) inframe->data[0])[(i * pixel_stride +
-                      j) * channels + 1])) - mean;
-      B = (gfloat) ((((guchar *) inframe->data[0])[(i * pixel_stride +
-                      j) * channels + 2])) - mean;
-      normalized = normalized + pow (R, 2);
-      normalized = normalized + pow (G, 2);
-      normalized = normalized + pow (B, 2);
-    }
-  }
-
-  variance = normalized / (float) (width * height * channels);
-  std = sqrt (variance);
-
-  for (i = 0; i < height; ++i) {
-    for (j = 0; j < width; ++j) {
-      ((gfloat *) outframe->data[0])[(i * width + j) * channels + 0] =
-          (((guchar *) inframe->data[0])[(i * pixel_stride + j) * channels +
-              0] - mean) / std;
-      ((gfloat *) outframe->data[0])[(i * width + j) * channels + 1] =
-          (((guchar *) inframe->data[0])[(i * pixel_stride + j) * channels +
-              1] - mean) / std;
-      ((gfloat *) outframe->data[0])[(i * width + j) * channels + 2] =
-          (((guchar *) inframe->data[0])[(i * pixel_stride + j) * channels +
-              2] - mean) / std;
-    }
-  }
-
-  return TRUE;
+  return gst_normalize_face (inframe, outframe, MODEL_CHANNELS);
 }
 
 static gboolean
@@ -263,25 +210,13 @@ gst_facenetv1_postprocess (GstVideoInference * vi, const gpointer prediction,
     gboolean * valid_prediction)
 {
   GstClassificationMeta *class_meta = (GstClassificationMeta *) meta_model;
-  GstDebugLevel level;
+  GstDebugLevel gst_debug_level = GST_LEVEL_LOG;
   GST_LOG_OBJECT (vi, "Postprocess");
 
-  class_meta->num_labels = predsize / sizeof (gfloat);
-  class_meta->label_probs =
-      g_malloc (class_meta->num_labels * sizeof (gdouble));
-  for (gint i = 0; i < class_meta->num_labels; ++i) {
-    class_meta->label_probs[i] = (gdouble) ((gfloat *) prediction)[i];
-  }
-
-  /* Only display vector if debug level >= 6 */
-  level = gst_debug_category_get_threshold (gst_facenetv1_debug_category);
-  if (level >= GST_LEVEL_LOG) {
-    for (gint i = 0; i < class_meta->num_labels; ++i) {
-      gfloat current = ((gfloat *) prediction)[i];
-      GST_LOG_OBJECT (vi, "Output vector element %i : (%f)", i, current);
-    }
-  }
+  gst_fill_classification_meta (class_meta, prediction, predsize);
 
+  gst_inference_print_embedding (vi, gst_facenetv1_debug_category, class_meta,
+      prediction, gst_debug_level);
   *valid_prediction = TRUE;
   return TRUE;
 }

ext/r2inference/gstfacenetv1.c

@@ -0,0 +1,178 @@
+/*
+ * GStreamer
+ * Copyright (C) 2019 RidgeRun
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Library General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Library General Public License for more details.
+ *
+ * You should have received a copy of the GNU Library General Public
+ * License along with this library; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 02111-1307, USA.
+ *
+ */
+
+/**
+ * SECTION:element-gstinceptionv1
+ *
+ * The inceptionv1 element allows the user to infer/execute a pretrained model
+ * based on the GoogLeNet (Inception v1 or Inception v2) architectures on 
+ * incoming image frames.
+ *
+ * <refsect2>
+ * <title>Example launch line</title>
+ * |[
+ * gst-launch-1.0 -v videotestsrc ! inceptionv1 ! xvimagesink
+ * ]|
+ * Process video frames from the camera using a GoogLeNet (Inception v1 or 
+ * Inception v2) model.
+ * </refsect2>
+ */
+
+#ifdef HAVE_CONFIG_H
+#include "config.h"
+#endif
+
+#include "gstinceptionv1.h"
+#include "gst/r2inference/gstinferencemeta.h"
+#include <string.h>
+#include "gst/r2inference/gstinferencepreprocess.h"
+#include "gst/r2inference/gstinferencepostprocess.h"
+#include "gst/r2inference/gstinferencedebug.h"
+
+GST_DEBUG_CATEGORY_STATIC (gst_inceptionv1_debug_category);
+#define GST_CAT_DEFAULT gst_inceptionv1_debug_category
+
+#define MEAN 128.0
+#define STD 1/128.0
+#define MODEL_CHANNELS 3
+
+static gboolean gst_inceptionv1_preprocess (GstVideoInference * vi,
+    GstVideoFrame * inframe, GstVideoFrame * outframe);
+static gboolean gst_inceptionv1_postprocess (GstVideoInference * vi,
+    const gpointer prediction, gsize predsize, GstMeta * meta_model,
+    GstVideoInfo * info_model, gboolean * valid_prediction);
+static gboolean gst_inceptionv1_start (GstVideoInference * vi);
+static gboolean gst_inceptionv1_stop (GstVideoInference * vi);
+
+enum
+{
+  PROP_0
+};
+
+/* pad templates */
+#define CAPS								\
+  "video/x-raw, "							\
+  "width=224, "							\
+  "height=224, "							\
+  "format={RGB, RGBx, RGBA, BGR, BGRx, BGRA, xRGB, ARGB, xBGR, ABGR}"
+
+static GstStaticPadTemplate sink_model_factory =
+GST_STATIC_PAD_TEMPLATE ("sink_model",
+    GST_PAD_SINK,
+    GST_PAD_REQUEST,
+    GST_STATIC_CAPS (CAPS)
+    );
+
+static GstStaticPadTemplate src_model_factory =
+GST_STATIC_PAD_TEMPLATE ("src_model",
+    GST_PAD_SRC,
+    GST_PAD_REQUEST,
+    GST_STATIC_CAPS (CAPS)
+    );
+
+struct _GstInceptionv1
+{
+  GstVideoInference parent;
+};
+
+struct _GstInceptionv1Class
+{
+  GstVideoInferenceClass parent;
+};
+
+/* class initialization */
+
+G_DEFINE_TYPE_WITH_CODE (GstInceptionv1, gst_inceptionv1,
+    GST_TYPE_VIDEO_INFERENCE,
+    GST_DEBUG_CATEGORY_INIT (gst_inceptionv1_debug_category, "inceptionv1", 0,
+        "debug category for inceptionv1 element"));
+
+static void
+gst_inceptionv1_class_init (GstInceptionv1Class * klass)
+{
+  GstElementClass *element_class = GST_ELEMENT_CLASS (klass);
+  GstVideoInferenceClass *vi_class = GST_VIDEO_INFERENCE_CLASS (klass);
+
+  gst_element_class_add_static_pad_template (element_class,
+      &sink_model_factory);
+  gst_element_class_add_static_pad_template (element_class, &src_model_factory);
+
+  gst_element_class_set_static_metadata (GST_ELEMENT_CLASS (klass),
+      "inceptionv1", "Filter",
+      "Infers incoming image frames using a pretrained GoogLeNet (Inception v1 or Inception v2) model",
+      "Carlos Rodriguez <[email protected]> \n\t\t\t"
+      "   Jose Jimenez <[email protected]> \n\t\t\t"
+      "   Michael Gruner <[email protected]> \n\t\t\t"
+      "   Mauricio Montero <[email protected]>");
+
+  vi_class->start = GST_DEBUG_FUNCPTR (gst_inceptionv1_start);
+  vi_class->stop = GST_DEBUG_FUNCPTR (gst_inceptionv1_stop);
+  vi_class->preprocess = GST_DEBUG_FUNCPTR (gst_inceptionv1_preprocess);
+  vi_class->postprocess = GST_DEBUG_FUNCPTR (gst_inceptionv1_postprocess);
+  vi_class->inference_meta_info = gst_classification_meta_get_info ();
+}
+
+static void
+gst_inceptionv1_init (GstInceptionv1 * inceptionv1)
+{
+}
+
+static gboolean
+gst_inceptionv1_preprocess (GstVideoInference * vi,
+    GstVideoFrame * inframe, GstVideoFrame * outframe)
+{
+  GST_LOG_OBJECT (vi, "Preprocess");
+  return gst_normalize (inframe, outframe, MEAN, STD, MODEL_CHANNELS);
+}
+
+static gboolean
+gst_inceptionv1_postprocess (GstVideoInference * vi, const gpointer prediction,
+    gsize predsize, GstMeta * meta_model, GstVideoInfo * info_model,
+    gboolean * valid_prediction)
+{
+  GstClassificationMeta *class_meta = (GstClassificationMeta *) meta_model;
+  GstDebugLevel gst_debug_level = GST_LEVEL_LOG;
+  GST_LOG_OBJECT (vi, "Postprocess");
+
+  gst_fill_classification_meta (class_meta, prediction, predsize);
+
+  gst_inference_print_highest_probability (vi, gst_inceptionv1_debug_category,
+      class_meta, prediction, gst_debug_level);
+
+  *valid_prediction = TRUE;
+  return TRUE;
+}
+
+static gboolean
+gst_inceptionv1_start (GstVideoInference * vi)
+{
+  GST_INFO_OBJECT (vi, "Starting Inception v1");
+
+  return TRUE;
+}
+
+static gboolean
+gst_inceptionv1_stop (GstVideoInference * vi)
+{
+  GST_INFO_OBJECT (vi, "Stopping Inception v1");
+
+  return TRUE;
+}

ext/r2inference/gstinceptionv1.c

@@ -0,0 +1,34 @@
+/*
+ * GStreamer
+ * Copyright (C) 2018 RidgeRun
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Library General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Library General Public License for more details.
+ *
+ * You should have received a copy of the GNU Library General Public
+ * License along with this library; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 02111-1307, USA.
+ *
+ */
+
+#ifndef _GST_INCEPTIONV1_H_
+#define _GST_INCEPTIONV1_H_
+
+#include <gst/r2inference/gstvideoinference.h>
+
+G_BEGIN_DECLS
+
+#define GST_TYPE_INCEPTIONV1 gst_inceptionv1_get_type ()
+G_DECLARE_FINAL_TYPE (GstInceptionv1, gst_inceptionv1, GST, INCEPTIONV1, GstVideoInference)
+
+G_END_DECLS
+
+#endif