Hello, all

maybe someone out there can throw in an idea to help me out with the following. I am hooking to an arbitrary D3D9 application, in most cases it is a game, and injecting my own code to mofify the behavior of the EndScene function. The backbuffer is copied into a surface which is set to point to a bitmap in a push source DirectShow filter. The filter samples the bitmaps at 25 fps and streams the video into an .avi file. There is a text overlay shown across the game's screen telling the user about a hot key combination that is supposed to stop gameplay capture. This overlay is not supposed to show up in the recoreded video. Everything works fast and beautiful except for one annoying fact. On a random occasion, a frame with the text overaly makes its way into the recoreded video. This is not a really desired artefact, the end user only wants to see his gameplay in the video and nothing else. I'd love to hear if anyone can share ideas of why this is happening. Here is the C# source code for the EndScene hook:

using System;
using SlimDX;
using SlimDX.Direct3D9;
using System.Diagnostics;
using DirectShowLib;
using System.Runtime.InteropServices;
[InterfaceType(ComInterfaceType.InterfaceIsIUnknown)]
[System.Security.SuppressUnmanagedCodeSecurity]
[Guid("EA2829B9-F644-4341-B3CF-82FF92FD7C20")]
public interface IScene
{
unsafe int PassMemoryPtr(void* ptr, bool noheaders);
int SetBITMAPINFO([MarshalAs(UnmanagedType.LPArray, SizeParamIndex = 1)]byte[] ptr, bool noheaders);
}
public class Class1
{
object _lockRenderTarget = new object();
public string StatusMess { get; set; }
Surface _renderTarget;
//points to image bytes
unsafe void* bytesptr;
//used to store headers AND image bytes
byte[] bytes;
IFilterGraph2 ifg2;
ICaptureGraphBuilder2 icgb2;
IBaseFilter push;
IBaseFilter compressor;
IScene scene;
IBaseFilter mux;
IFileSinkFilter sink;
IMediaControl media;
bool NeedRunGraphInit = true;
bool NeedRunGraphClean = true;
DataStream s;
DataRectangle dr;
unsafe int EndSceneHook(IntPtr devicePtr)
{
int hr;

using (Device device = Device.FromPointer(devicePtr))
{
try
{
lock (_lockRenderTarget)
{

bool TimeToGrabFrame = false;
//....
//logic based on elapsed milliseconds deciding if it is time to grab another frame

if (TimeToGrabFrame)
{

//First ensure we have a Surface to render target data into
//called only once
if (_renderTarget == null)
{

//Create offscreen surface to use as copy of render target data
using (SwapChain sc = device.GetSwapChain(0))
{

//Att: created in system memory, not in video memory
_renderTarget = Surface.CreateOffscreenPlain(device, sc.PresentParameters.BackBufferWidth, sc.PresentParameters.BackBufferHeight, sc.PresentParameters.BackBufferFormat, Pool.SystemMemory);

} //end using
} // end if

using (Surface backBuffer = device.GetBackBuffer(0, 0))
{
//The following line is where main action takes place:
//Direct3D 9 back buffer gets copied to Surface _renderTarget,
//which has been connected by references to DirectShow's
//bitmap capture filter
//Inside the filter ( code not shown in this listing) the bitmap is periodically
//scanned to create a streaming video.
device.GetRenderTargetData(backBuffer, _renderTarget);

if (NeedRunGraphInit) //ran only once
{
ifg2 = (IFilterGraph2)new FilterGraph();
icgb2 = (ICaptureGraphBuilder2)new CaptureGraphBuilder2();
icgb2.SetFiltergraph(ifg2);
push = (IBaseFilter) new PushSourceFilter();
scene = (IScene)push;

//this way we get bitmapfile and bitmapinfo headers
//ToStream is slow, but run it only once to get the headers
s = Surface.ToStream(_renderTarget, ImageFileFormat.Bmp);
bytes = new byte[s.Length];

s.Read(bytes, 0, (int)s.Length);
hr = scene.SetBITMAPINFO(bytes, false);
//we just supplied the header to the PushSource
//filter. Let's pass reference to
//just image bytes from LockRectangle

dr = _renderTarget.LockRectangle(LockFlags.None);
s = dr.Data;
Result r = _renderTarget.UnlockRectangle();
bytesptr = s.DataPointer.ToPointer();
hr = scene.PassMemoryPtr(bytesptr, true);

//continue building graph
ifg2.AddFilter(push, "MyPushSource");

icgb2.SetOutputFileName(MediaSubType.Avi, "C:\foo.avi", out mux, out sink);

icgb2.RenderStream(null, null, push, null, mux);

media = (IMediaControl)ifg2;

media.Run();

NeedRunGraphInit = false;
NeedRunGraphClean = true;
StatusMess = "now capturing, press shift-F11 to stop";

} //end if

} // end using backbuffer
} // end if Time to grab frame

} //end lock
} // end try

//It is usually thrown when the user makes game window inactive
//or it is thrown deliberately when time is up, or the user pressed F11 and
//it resulted in stopping a capture.
//If it is thrown for another reason, it is still a good
//idea to stop recording and free the graph
catch (Exception ex)
{
//..
//stop the DirectShow graph and cleanup

} // end catch
//draw overlay
using (SlimDX.Direct3D9.Font font = new SlimDX.Direct3D9.Font(device, new System.Drawing.Font("Times New Roman", 26.0f, FontStyle.Bold)))
{
font.DrawString(null, StatusMess, 20, 100, System.Drawing.Color.FromArgb(255, 255, 255, 255));
}

return device.EndScene().Code;

} // end using device

} //end EndSceneHook

As it happens sometimes, I finally found an answer to this question myself, if anyone is interested. It turned out that backbuffer in some Direct3D9 apps is not necessarily refreshed each time the hooked EndScene is called. Hence, occasionally the backbuffer with the text overlay from the previous EndScene hook call was passed to the DirectShow source filter responsible for collecting input frames. I started stamping each frame with a tiny overlay with known RGB values and checking if it was still present before passing it to the DirectShow filter. If it was there, the previously cached frame was passed instead of the current one. This approach effectively removed the text overlay from the video recorded in the DirectShow graph.