Encode YUV420 buffer with appsrc

[Antonio Ospite]

On Tue, 24 Oct 2017 02:39:41 -0700 (MST)
pchaurasia <pchaurasia at gameloreinc.com> wrote:

> Hi Antonio,
> 
> 1. When I set image width to 2048, I do get better looking video. However,
> It seems colors are not perfectly right, in the video. I think I am 50%
> there.

This is because the data format is not actually I420 but YV12, the U and
V plane are switched.

Setting the width equal to the stride helped to understand this.

> 2. I dumped the image as raw pixel image of size 2048*1080*1.5. In YUV
> Planar format the way you described above. I am attaching the image.

A very handy way to figure out the raw data format is to use the
rawvideoparse element, for example I can correctly interpret the frame
you sent with the following script:

-----------------------------------------------------------------------
#!/bin/sh

set -e
set -x

WIDTH=1920
HEIGHT=1080

STRIDE=2048

STRIDE1=$STRIDE
STRIDE2=$(($STRIDE / 2))
STRIDE3=$(($STRIDE / 2))

OFFSET1=0
OFFSET2=$(($STRIDE * $HEIGHT))
OFFSET3=$(($STRIDE * $HEIGHT + ($STRIDE / 2) * ($HEIGHT / 2))) 

gst-launch-1.0 filesrc location=frame0065.jpg ! \
  rawvideoparse \
    width=$WIDTH \
    height=$HEIGHT \
    format=yv12 \
    plane-strides="<$STRIDE1,$STRIDE2,$STRIDE3>" \
    plane-offsets="<$OFFSET1,$OFFSET2,$OFFSET3>" ! \
  videoconvert ! pngenc ! filesink location=out.png
-----------------------------------------------------------------------

> 3. I am attaching the code which will load this image and try to encode.
> This code also displays the image before encoding. WHen I run , I see that
> pre-encoded image as displayed by imshow() give right colors - however - the
> encoded image does not.

To set the offsets you have to calculate the planes sizes in bytes like
above; then set the right pixelformat and you will have an almost correct
image. I say _almost_ because there is still a border on the right.

You can fix that by differentiating again between the width of the final
image and the stride of the raw data.

int stride = 2048;
int width = 1920;

and use these variables appropriately, as you were doing before.

I didn't do that in the patch below to keep the changes at a minimum.

$ diff -pruN main3.cpp.orig main3.cpp
--- main3.cpp.orig	2017-10-24 16:57:14.209146385 +0200
+++ main3.cpp	2017-10-24 17:38:22.660606798 +0200
@@ -47,7 +47,7 @@ typedef struct _CustomData {
 
 int width             = 2048;
 int height            = 1080;
-static int fno        = 60;
+static int fno        = 65;
 unsigned char         *imgdata;
 int dx                = 0;
 int dy                = 0;
@@ -64,8 +64,13 @@ static void start_feed (GstElement *sour
   gsize            m_offset[3];
   gint             m_stride[3];
 
+  // at the beginning of the data
+  m_offset[0] = 0;
+  // after the first plane
+  m_offset[1] = width * height;
+  // after the first and second plane
+  m_offset[2] = width * height + (width / 2) * (height / 2); // or: width * height * 1.25
 
-  m_offset[0] = m_offset[1] = m_offset[2] = 0;
   m_stride[0] = width; 
   m_stride[1] = width/2; 
   m_stride[2] = width/2;
@@ -109,7 +114,7 @@ static void start_feed (GstElement *sour
 #endif
 
     m_pgstBuffer           = gst_buffer_new_wrapped_full( (GstMemoryFlags)0, (gpointer)(inputImgGray.data), size, 0, size, NULL, NULL );
-    m_pgstVideoMeta        = gst_buffer_add_video_meta_full(m_pgstBuffer,GST_VIDEO_FRAME_FLAG_NONE, GST_VIDEO_FORMAT_I420, width,height, 3, m_offset, m_stride );
+    m_pgstVideoMeta        = gst_buffer_add_video_meta_full(m_pgstBuffer,GST_VIDEO_FRAME_FLAG_NONE, GST_VIDEO_FORMAT_YV12, width,height, 3, m_offset, m_stride );
 
     //ref buffer to give copy to appsrc
     gst_buffer_ref(m_pgstBuffer);
@@ -176,7 +181,7 @@ main (int argc, char * argv[])
   data.app_source             = gst_element_factory_make ("appsrc", "audio_source");
   //data.app_sink               = gst_element_factory_make ("appsink", "app_sink");
   data.m_pvideoConvert        = gst_element_factory_make("autovideoconvert", "aa-videoconvert");
-  data.m_pencoder             = gst_element_factory_make("omxh265enc", "aa-videoencoder");
+  data.m_pencoder             = gst_element_factory_make("x264enc", "aa-videoencoder");
   data.m_pmux                 = gst_element_factory_make("matroskamux", "aa-mux");
   data.m_pfsink               = gst_element_factory_make("filesink", "aa-filesink");
   data.sourceid               = 0; 
@@ -192,11 +197,11 @@ main (int argc, char * argv[])
   }
 
 
-  gst_video_info_set_format(&info, GST_VIDEO_FORMAT_I420, width, height);
+  gst_video_info_set_format(&info, GST_VIDEO_FORMAT_YV12, width, height);
   caps                 = gst_video_info_to_caps(&info); 
 
   caps = gst_caps_new_simple ("video/x-raw",
-			"format",G_TYPE_STRING,"I420",
+			"format",G_TYPE_STRING,"YV12",
 			"width", G_TYPE_INT, width,
 			"height", G_TYPE_INT, height,
 			"framerate", GST_TYPE_FRACTION, 30, 1,


Ciao ciao,
   Antonio

-- 
Antonio Ospite
https://ao2.it
https://twitter.com/ao2it

A: Because it messes up the order in which people normally read text.
   See http://en.wikipedia.org/wiki/Posting_style
Q: Why is top-posting such a bad thing?